Intelligent drunk driving monitoring method and system

Abstract

The invention relates to an intelligent drunk driving monitoring method and system, and the system comprises an alcohol sensor 1, a voice collection module 2, an image collection module 3, a dynamic detection module 4, a positioning module 5, a main control chip 6, and a response module 7. The alcohol sensor 1 is disposed in a breathing region of a driver, and the voice collection module 2 is disposed in a front cabin of a vehicle. The image acquisition module 3 is arranged at the front part of a vehicle, the dynamic detection module 4 is arranged in the vehicle, the positioning module 5 is arranged in the vehicle, and the main control chip 6 is used for constructing alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information and geographic position information in a preset time window to obtain a behavior state evolution sequence; and the response module 7 is connected with the output end of the main control chip 6 and is used for executing response operation according to the event identification result generated by the main control chip 6. The method has the effect of improving the identification precision of the potential drunk driving behavior.

Description

Technical Field

[0001] This invention relates to the technical field of traffic safety monitoring, and in particular to an intelligent method and system for detecting drunk driving. Background Technology

[0002] Currently, several methods are commonly used to detect drunk driving. One common method is to use a single alcohol sensor, typically installed inside the vehicle, to determine whether the driver is intoxicated by detecting the alcohol concentration in their breath. This method is relatively simple to operate and has a low cost. Another method relies on traffic police setting up checkpoints at specific locations and using specialized alcohol testing equipment to test passing drivers. This method can directly determine whether a driver is intoxicated, but it requires a significant investment of manpower and resources and can only be conducted at specific locations and times. In addition, some areas analyze vehicle trajectories and behaviors using surveillance cameras, such as observing whether vehicles frequently exhibit abnormal driving patterns. However, this method can only serve as a supplementary means and is difficult to accurately determine whether a driver is intoxicated.

[0003] However, existing methods for detecting drunk driving have significant shortcomings. A single alcohol sensor can only detect the alcohol concentration in a driver's breath, failing to consider other behavioral characteristics and potentially leading to missed or false detections. Police checkpoints have limitations, unable to monitor drunk driving on the road in real-time and comprehensively. Surveillance cameras, while analyzing vehicle trajectories and behavior, lack sufficient accuracy and reliability, failing to detect drunk driving promptly and effectively, thus falling short of the growing demands for road safety. Summary of the Invention

[0004] To improve the accuracy of identifying potential drunk driving behavior, this application provides an intelligent drunk driving monitoring method and system.

[0005] The above-mentioned objective of this application is achieved through the following technical solution: An intelligent drunk driving monitoring system includes an alcohol sensor, a voice acquisition module, an image acquisition module, a dynamic detection module, a positioning module, a main control chip, and a response module, wherein: An alcohol sensor is placed in the driver's breathing area. The output of the alcohol sensor is used to collect the driver's alcohol concentration information and output the alcohol concentration information to the first input of the main control chip. The voice acquisition module is located inside the vehicle's front cabin. The output of the voice acquisition module is used to acquire the driver's voice signal, extract voice feature information, and transmit the voice feature information to the second input of the main control chip. The image acquisition module is located at the front of the vehicle. The image acquisition module is used to acquire image information in front of the vehicle, and to identify the image information in front of the vehicle based on the lane recognition algorithm to obtain lane departure feature information. The lane departure feature information is then transmitted to the third input terminal of the main control chip. The dynamic detection module is located inside the vehicle. It is used to acquire the vehicle's acceleration information, process the acceleration information to obtain dynamic feature information, including emergency braking behavior feature information, lateral swaying behavior feature information and collision behavior feature information, and transmit the dynamic feature information to the fourth input terminal of the main control chip. The positioning module is located inside the vehicle. The positioning module is used to obtain the vehicle's geographical location information and transmit the geographical location information to the fifth input terminal of the main control chip. The main control chip is used to construct information on alcohol concentration, voice features, lane departure, dynamic features and geographic location within a preset time window to obtain a behavioral state evolution sequence, and to perform trend analysis on the behavioral state evolution sequence to generate event recognition results. The response module is located inside the vehicle and is connected to the output of the main control chip. The response module is used to perform response operations based on the event recognition results generated by the main control chip.

[0006] By adopting the above technical solution, it is possible to achieve multi-source synchronous collection and fusion analysis of driver's alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information, and geographical location information. This enables the construction of a behavioral state evolution sequence reflecting changes in driving behavior, and the generation of event recognition results by combining trend analysis. This allows for real-time and continuous comprehensive judgment of drunk driving risk during vehicle operation, avoiding the problems of missed or false detections caused by relying solely on a single alcohol concentration test. Furthermore, it is not limited by the time and location of traffic police checkpoints, and it overcomes the shortcomings of insufficient accuracy in analyzing driving trajectories by surveillance cameras. This achieves intelligent drunk driving monitoring that is online throughout the entire process, which helps to promptly detect and deal with potential drunk driving behaviors and improve road traffic safety.

[0007] In a preferred embodiment, this application can be further configured such that the main control chip includes a feature sequence construction unit and a trend judgment unit, wherein: The feature sequence construction unit is used to perform time-series sampling of alcohol concentration information, voice feature information, lane departure feature information and dynamic feature information within a preset time window, construct a behavior state evolution sequence containing multiple time nodes, and send the behavior state evolution sequence to the trend judgment unit. The trend judgment unit is used to calculate the time change gradient, fluctuation amplitude or persistence offset index of each set of feature information based on multiple sets of feature information in the behavioral state evolution sequence, and to determine whether there is a risk trend of abnormal increase of a single feature information or synergistic increase of multiple feature information. If a risk trend is determined to exist, an event recognition result is generated at the output of the main control chip.

[0008] By adopting the above technical solution, multi-dimensional temporal sampling and unified construction of alcohol concentration information, voice feature information, lane departure feature information, and dynamic feature information can be performed within a fixed time window to form a behavioral state evolution sequence that reflects the changes in driving behavior over time. Furthermore, the trend judgment unit is used to quantitatively analyze the change gradient, fluctuation amplitude, and continuous deviation of multiple sets of feature information, thereby accurately identifying the risk trend of single feature anomalies or multiple features increasing synergistically. This enables early detection and intelligent warning of potential drunk driving risks, improving the real-time performance and reliability of drunk driving monitoring.

[0009] In a preferred embodiment, this application can be further configured such that the response module includes a local alert unit and a remote communication unit, wherein: The local alert unit is located in the cockpit and is connected to the output of the main control chip. When the event recognition result indicates that the driver is distracted or has a tendency to drive while fatigued, the local alert unit is used to trigger the voice broadcast device and the warning light device to output voice prompts and visual flashing signals. The remote communication unit is connected to the output of the main control chip and is used to send alarm information to preset emergency contacts via wireless communication when the event recognition result indicates that the alcohol level is too high or there is a high-risk driving trend. The alarm information includes the event type, alcohol concentration value, vehicle geographical location information and video clips.

[0010] By adopting the above technical solution, when a tendency for inattention or fatigued driving is detected, the driver can be immediately alerted in real time in the cockpit with voice prompts and visual flashing signals, prompting the driver to adjust their driving status in time. At the same time, when an excessive alcohol level or a high-risk driving trend is detected, the event type, alcohol concentration value, vehicle geographical location information and video clips are sent to a preset emergency contact via wireless communication, realizing the combination of local instant reminders and remote information push, thereby forming a multi-layered risk response mechanism.

[0011] A smart drunk driving detection method, comprising: The alcohol sensor collects the driver's alcohol concentration information and sends the information to the main control chip. The driver's voice signal is collected by the voice acquisition module, the voice feature information is extracted, and the voice feature information is sent to the main control chip. The image acquisition module acquires image information in front of the vehicle, analyzes the image information based on the lane recognition algorithm, obtains lane departure feature information, and sends the lane departure feature information to the main control chip. The vehicle's acceleration information is acquired through a dynamic detection module, processed to obtain dynamic feature information, which includes emergency braking behavior feature information, lateral swaying behavior feature information, and collision behavior feature information. The dynamic feature information is then sent to the main control chip. The vehicle's geographical location information is obtained through the positioning module and then sent to the main control chip. The main control chip constructs a behavioral state evolution sequence from alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information and geographical location information within a preset time window, performs trend analysis on the behavioral state evolution sequence, determines whether there is an increasing risk trend, and generates event recognition results. The event identification results are output to the response module, which then performs the corresponding response operation based on the level of the event identification results.

[0012] By adopting the above technical solution, alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information, and geographical location information can be collected and comprehensively analyzed simultaneously within the same time window. This enables dynamic monitoring of the driver's alcohol status and multi-dimensional driving behavior, and accurately identifies the risk increase trend based on the trend analysis results of the behavioral state evolution sequence. Thus, after the event identification result is generated, local prompts or remote alarms can be triggered in a timely manner according to the risk level, which helps to improve the timeliness of detection and handling efficiency of drunk driving and high-risk driving behaviors.

[0013] In a preferred embodiment, this application can be further configured to: analyze image information based on a lane recognition algorithm to obtain lane departure feature information, including: The image acquisition module acquires a continuous sequence of image frames in front of the vehicle, and performs edge enhancement processing on each image frame to obtain an image frame with enhanced edges. The edge detection algorithm is used to process the edge-enhanced image frame to extract the left lane line boundary and the right lane line boundary; Based on the left lane line boundary and the right lane line boundary, the position of the lane center line in each image frame is estimated by a multi-frame curve fitting method to obtain the lane center line coordinate sequence. Based on the vehicle contour information in the edge-enhanced image frame, the coordinates of the vehicle center point are extracted to obtain the vehicle center point coordinate sequence; The vehicle center point coordinate sequence is compared with the lane center line coordinate sequence in the corresponding image frame to calculate the lateral offset distance of the vehicle relative to the lane center line, thus obtaining the lateral offset distance sequence. In the lateral offset distance sequence, it is determined whether the lateral offset distances corresponding to multiple consecutive image frames all exceed a preset offset threshold. If the lateral offset distances corresponding to multiple consecutive image frames all exceed the preset offset threshold, the turn signal status information in the corresponding image frames is combined to determine whether there is a lateral offset behavior when the turn signal is not turned on. When it is determined that there is a lateral offset behavior when the turn signal is not turned on, lane departure feature information is generated.

[0014] By adopting the above technical solution, the relationship between lane lines and vehicle position can be accurately calculated based on a continuous image frame sequence during vehicle operation. Combined with the judgment of lateral offset distance and the analysis of turn signal status information, lane departure behavior when the turn signal is not turned on can be accurately identified, thereby generating lane departure feature information and providing accurate and stable input data support for subsequent risk trend analysis and event recognition.

[0015] In a preferred embodiment, this application can be further configured as follows: if the lateral offset distances corresponding to multiple consecutive image frames all exceed a preset offset threshold, then, based on the turn signal status information in the corresponding image frames, it is determined whether there is a lateral offset behavior when the turn signal is not activated, including: Interpolation smoothing is performed on the vehicle center point coordinate sequence extracted from multiple consecutive image frames, and the position change trend of the vehicle center point coordinate sequence in the horizontal direction is calculated to obtain the vehicle driving direction change characteristics. The vehicle's direction of travel change features are matched with turn signal status information extracted from multiple consecutive image frames to determine whether there is a turn signal activation signal during the period of continuous lateral position shift. If no turn signal is detected during the period of continuous lateral displacement, and the vehicle's direction of travel changes in accordance with the pattern of continuous lateral movement, then it is determined that there is lateral displacement behavior without the turn signal being activated.

[0016] By adopting the above technical solution, it is possible to combine the interpolation and smoothing results of the vehicle center point coordinate sequence with the trend of horizontal position change during vehicle driving, and comprehensively judge the correspondence between the continuous lateral position deviation and the turn signal status. Thus, when no turn signal is detected and the vehicle driving direction change characteristics conform to the continuous lateral movement pattern, the lateral deviation behavior under the condition that the turn signal is not turned on can be accurately identified, thereby improving the accuracy and stability of lane departure feature information recognition.

[0017] In a preferred embodiment, this application can be further configured to: process the acceleration information to obtain dynamic feature information, which includes emergency braking behavior feature information, lateral swaying behavior feature information, and collision behavior feature information, including: At each time sampling point, the linear acceleration components in the forward and backward directions are extracted from the acceleration information. It is then determined whether the linear acceleration components in the forward and backward directions are lower than the emergency braking judgment threshold. If the linear acceleration components in the forward and backward directions are lower than the emergency braking judgment threshold, the changes in the linear acceleration components in the forward and backward directions are used as emergency braking behavior feature information. Extract the corresponding lateral acceleration components in the left and right directions from the acceleration information at multiple consecutive time sampling points, determine whether there is an alternation between positive and negative directions in the lateral acceleration components, and if the amplitude of the alternation exceeds the lateral swaying judgment threshold, then the change of the lateral acceleration components is used as the lateral swaying behavior feature information. At each time sampling point, the rate of change of the acceleration vector is calculated based on the acceleration information. It is then determined whether the rate of change of the acceleration vector exceeds the collision determination threshold. If it exceeds the collision determination threshold, the change in the rate of change of the acceleration vector at that time sampling point is used as the collision behavior feature information.

[0018] By adopting the above technical solution, it is possible to determine and analyze the linear acceleration components in the front-to-back direction, the lateral acceleration components in the left-to-right direction, and the rate of change of the acceleration vector based on acceleration information. This enables accurate identification of emergency braking behavior characteristics, lateral swaying behavior characteristics, and collision behavior characteristics under different driving conditions, achieving multi-dimensional and accurate extraction of vehicle dynamic feature information and improving the reliability and timeliness of event recognition results.

[0019] In a preferred embodiment, this application can be further configured such that: the main control chip constructs a behavioral state evolution sequence based on alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information, and geographical location information within a preset time window, including: Within a preset time window, extract the alcohol concentration value from the alcohol concentration information, the speech rate and tone parameters from the speech feature information, the lateral offset distance sequence from the lane departure feature information, the emergency braking behavior feature information, the lateral swaying behavior feature information and the collision behavior feature information from the dynamic feature information, and the location coordinates from the geographic location information. According to each sampling point in the preset time window, the alcohol concentration value, speech rate parameter and tone parameter, lateral offset distance, emergency braking behavior feature information, lateral swaying behavior feature information, collision behavior feature information and position coordinates are synchronously aligned to construct a time series state vector set; Behavioral modeling is performed on the time series state vector set to extract behavioral change features and generate a behavioral state evolution sequence that represents the trend of driving behavior changes.

[0020] By adopting the above technical solution, it is possible to simultaneously fuse alcohol concentration values, voice feature parameters, lane departure information, dynamic behavior feature information and location coordinates from multiple sources under a unified time reference, and perform behavior modeling based on the time series state vector set to form a behavior state evolution sequence that can reflect the changing trend of driving behavior, thereby providing complete, continuous and highly correlated basic data support for subsequent trend analysis and risk identification.

[0021] In a preferred example, this application can be further configured to: perform trend analysis on the behavioral state evolution sequence, determine whether there is an increasing risk trend, and generate event identification results, including: Risk scores are assigned to the behavioral change characteristics at each time sampling point in the behavioral state evolution sequence to obtain a risk score time series. The trend fitting process within a sliding window is performed on the scoring time series, the slope of the score change is calculated, and it is determined whether the slope of the score change exceeds the risk trend judgment threshold. If the slope of the score change is positive and exceeds the preset risk trend judgment threshold, then it is determined that there is an increasing risk trend. Given the established trend of increasing risk, the corresponding event identification results are generated by combining abnormal feature information in the behavioral state evolution sequence.

[0022] By adopting the above technical solution, the dynamic evolution of driving behavior risk can be quantified by utilizing the trend changes in the scoring time series. When the slope of the score change exceeds the preset risk trend judgment threshold, the risk increase trend can be identified in a timely manner. Combined with the abnormal feature information in the behavioral state evolution sequence, event recognition results are generated, thereby realizing trend-based risk judgment and event recognition processing based on multi-source behavioral data.

[0023] In a preferred example, this application can be further configured to include: anomaly characteristic information, including: In the alcohol concentration information, if the alcohol concentration values ​​at multiple consecutive time sampling points exceed the preset alcohol threshold, the corresponding changes in alcohol concentration values ​​will be used as alcohol concentration anomaly information in the abnormal feature information. In speech feature information, if the speech rate parameter is higher or lower than the normal speech rate range, or the tone parameter fluctuates within a preset time, the speech rate parameter and tone parameter at the corresponding time sampling point will be regarded as speech behavior abnormal information in the abnormal feature information. In lane departure feature information, if the lateral offset distance in multiple consecutive image frames exceeds the offset threshold and there is no turn signal status information, then the lateral offset distance sequence and the turn signal status combination information are used as lane departure anomaly information in the anomaly feature information. If emergency braking behavior feature information, lateral swaying behavior feature information, or collision behavior feature information is detected within a preset time window in the dynamic feature information, then the emergency braking behavior feature information, lateral swaying behavior feature information, or collision behavior feature information will be regarded as dynamic driving abnormal information in the abnormal feature information. In geographic location information, if the driving path corresponding to the location coordinates of multiple time sampling points contains dangerous area markers or high-risk road segment labels, then the geographic location information will be marked as high-risk area information in the abnormal feature information.

[0024] By adopting the above technical solution, clear anomaly judgment conditions can be set for alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information and geographical location information respectively. When the corresponding conditions are met, the relevant data can be extracted and classified as abnormal feature information, thereby ensuring that the source of abnormal feature information is traceable and the judgment criteria are quantifiable, which facilitates multi-dimensional accurate correlation judgment and comprehensive analysis in subsequent risk trend analysis and event identification.

[0025] In summary, this application includes at least one of the following beneficial technical effects: 1. It can simultaneously collect and fuse driver alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information and geographical location information from multiple sources, construct a behavioral state evolution sequence reflecting changes in driving behavior, and generate event recognition results by combining trend analysis. This enables real-time and continuous comprehensive judgment of drunk driving risk during vehicle operation, avoiding the problem of missed or false detection caused by relying solely on a single alcohol concentration test. At the same time, it is not limited by the time and location of traffic police checkpoints, and it also overcomes the shortcomings of insufficient accuracy in analyzing driving trajectories by surveillance cameras. It achieves intelligent drunk driving monitoring online throughout the entire process, which helps to promptly detect and deal with potential drunk driving behavior and improve the level of road traffic safety. 2. It can simultaneously collect and comprehensively analyze alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information and geographical location information within the same time window, realize dynamic monitoring of driver's alcohol status and multi-dimensional driving behavior, and accurately identify the risk increase trend based on the trend analysis results of the behavior state evolution sequence. After the event identification result is generated, it can promptly trigger local prompts or remote alarms and other response operations according to the risk level, which helps to improve the timeliness of detection and handling efficiency of drunk driving and high-risk driving behavior. 3. It can quantify the dynamic evolution of driving behavior risks by utilizing the trend changes in the scoring time series, and promptly identify the risk increase trend when the slope of the score change exceeds the preset risk trend judgment threshold. It can also generate event recognition results by combining abnormal feature information in the behavioral state evolution sequence, thereby realizing trend-based risk judgment and event recognition processing based on multi-source behavioral data. Attached Figure Description

[0026] Figure 1 This is a flowchart of an intelligent drunk driving monitoring system according to one embodiment of this application.

[0027] Figure 2 This is a flowchart of an intelligent drunk driving detection method according to one embodiment of this application; Figure 3 This is a flowchart illustrating the implementation of step S30 in an intelligent drunk driving monitoring method according to an embodiment of this application. Figure 4 This is a flowchart illustrating the implementation of step S306 in an intelligent drunk driving monitoring method according to an embodiment of this application.

[0028] Explanation of reference numerals in the attached diagram: 1. Alcohol sensor; 2. Voice acquisition module; 3. Image acquisition module; 4. Dynamic detection module; 5. Positioning module; 6. Main control chip; 7. Response module. Detailed Implementation

[0029] The present application will be further described in detail below with reference to the accompanying drawings.

[0030] In one embodiment, such as Figure 1 As shown, this application discloses an intelligent drunk driving monitoring system, which includes an alcohol sensor 1, a voice acquisition module 2, an image acquisition module 3, a dynamic detection module 4, a positioning module 5, a main control chip 6, and a response module 7, wherein: Alcohol sensor 1 is placed in the driver's breathing area. The output of alcohol sensor 1 is used to collect the driver's alcohol concentration information and output the alcohol concentration information to the first input of the main control chip 6. The voice acquisition module 2 is located inside the front cabin of the vehicle. The output of the voice acquisition module 2 is used to acquire the driver's voice signal, extract voice feature information, and transmit the voice feature information to the second input of the main control chip 6. The image acquisition module 3 is located at the front of the vehicle. The image acquisition module 3 is used to acquire image information in front of the vehicle, and to identify the image information in front of the vehicle based on the lane recognition algorithm to obtain lane departure feature information. The lane departure feature information is then transmitted to the third input terminal of the main control chip 6. The dynamic detection module 4 is installed inside the vehicle. The dynamic detection module 4 is used to acquire the vehicle's acceleration information, process the acceleration information to obtain dynamic feature information, including emergency braking behavior feature information, lateral swaying behavior feature information and collision behavior feature information, and transmit the dynamic feature information to the fourth input terminal of the main control chip 6. The positioning module 5 is installed inside the vehicle. The positioning module 5 is used to obtain the geographical location information of the vehicle and transmit the geographical location information to the fifth input terminal of the main control chip 6. The main control chip 6 is used to construct information on alcohol concentration, voice features, lane departure features, dynamic features and geographic location within a preset time window, obtain a behavioral state evolution sequence, perform trend analysis on the behavioral state evolution sequence, and generate event recognition results. The response module 7 is located inside the vehicle and is connected to the output of the main control chip 6. The response module 7 is used to perform response operations based on the event recognition results generated by the main control chip 6.

[0031] Specifically, the alcohol sensor 1 is located in the driver's breathing area. It uses a gas sensing element to detect the concentration of ethanol molecules in the breath gas through optical or chemical reaction. The detected electrical signal is converted into a digital signal to form alcohol concentration information, which is then transmitted to the first input terminal of the main control chip 6 via a wired signal interface; (2) The voice acquisition module 2 is located inside the vehicle's front cabin. It collects the driver's voice signal through a pickup unit. After analog signal conditioning and analog-to-digital conversion, it extracts voice feature information, which includes speech rate parameters and pitch parameters obtained through acoustic analysis algorithms. The voice feature information is then transmitted to the second input terminal of the main control chip 6; (3) The image acquisition module Block 3 is located at the front of the vehicle. It collects image information in front of the vehicle through an imaging sensor, and performs pixel-level segmentation and lane line fitting on the image information using a lane recognition algorithm to obtain lane departure feature information. The lane departure feature information includes lateral offset distance and offset direction data, and transmits the lane departure feature information to the third input terminal of the main control chip 6; (4) The dynamic detection module 4 is located inside the vehicle. It collects acceleration information during vehicle operation through a triaxial acceleration sensor, decomposes and analyzes the acceleration information in the time series, identifies the change pattern of longitudinal acceleration that meets the emergency braking judgment condition to generate emergency braking behavior feature information, and identifies lateral acceleration. The system generates lateral sway behavior feature information by alternating patterns exceeding the lateral sway judgment threshold within a preset time window. When the instantaneous acceleration peak and impact waveform characteristics meet the collision judgment conditions, collision behavior feature information is generated, and the dynamic feature information is transmitted to the fourth input terminal of the main control chip 6; (5) The positioning module 5 is installed inside the vehicle and receives positioning signals transmitted by multiple satellites through the satellite positioning receiving unit. It combines timing information and three-dimensional coordinate calculation algorithm to obtain the vehicle's geographical location information, which includes location coordinates and timestamps, and transmits the geographical location information to the fifth input terminal of the main control chip 6; (6) The main control chip 6 generates lateral sway behavior feature information by alternating patterns exceeding the lateral sway judgment threshold within a preset time window. When the instantaneous acceleration peak and impact waveform characteristics meet the collision judgment conditions, collision behavior feature information is generated, and the dynamic feature information is transmitted to the fourth input terminal of the main control chip 6; (7) The positioning module 5 is installed inside the vehicle and receives positioning signals transmitted by multiple satellites through the satellite positioning receiving unit. It combines timing information and three-dimensional coordinate calculation algorithm to obtain the vehicle's geographical location information, which includes location coordinates and timestamps, and transmits the geographical location information to the fifth input terminal of the main control chip 6; (8) The main control chip 6 generates lateral sway behavior feature information by alternating patterns exceeding the lateral sway judgment threshold within a preset time window. Concentration information, voice feature information, lane departure feature information, dynamic feature information, and geographic location information are synchronously aligned according to time sampling points to construct a time series state vector set. Based on the time series state vector set, behavioral evolution modeling is performed to identify the fluctuation patterns and co-occurrence relationships of various feature information within a preset time window to generate a behavioral state evolution sequence. Then, trend analysis is performed on the behavioral state evolution sequence, and risk scores are assigned to feature information at each time point through evolution scoring rules. The slope of score change and risk trend judgment threshold are combined to determine whether there is a risk increase trend. If there is a risk increase trend, the corresponding abnormal feature information is extracted to generate event recognition results.(7) The response module 7 is connected to the output of the main control chip 6, and controls the execution unit to trigger corresponding response operations based on the event recognition results, including triggering the vehicle-mounted audible and visual alarm, sending control signals to the vehicle control bus to limit vehicle start-up or reduce vehicle speed, or sending alarm information to an external management platform via the wireless communication module for remote handling.

[0032] Furthermore, acoustic analysis algorithms refer to... A three-axis accelerometer refers to... A satellite positioning receiver unit refers to... In one embodiment, the main control chip 6 includes a feature sequence construction unit 61 and a trend judgment unit 62, wherein: The feature sequence construction unit 61 is used to perform time-series sampling of alcohol concentration information, voice feature information, lane departure feature information and dynamic feature information within a preset time window, construct a behavior state evolution sequence containing multiple time nodes, and send the behavior state evolution sequence to the trend judgment unit 62. The trend judgment unit 62 is used to calculate the time change gradient, fluctuation amplitude or continuous offset index of each set of feature information based on multiple sets of feature information in the behavioral state evolution sequence, and to determine whether there is a risk trend of abnormal rise of a single feature information or synergistic rise of multiple feature information. If it is determined that there is a risk trend, it generates an event recognition result at the output of the main control chip 6.

[0033] Specifically, the feature sequence construction unit 61 performs temporal sampling on alcohol concentration information, speech feature information, lane departure feature information, and dynamic feature information within a preset time window. Temporal sampling refers to discretizing and recording feature information on the time axis at fixed time intervals. Each time sampling point forms a state vector containing the alcohol concentration value from the alcohol concentration information, the speech rate and pitch parameters from the speech feature information, the lateral deviation distance from the lane departure feature information, and the emergency braking behavior feature information, lateral swaying behavior feature information, and collision behavior feature information from the dynamic feature information. The state vector is a vector composed of multiple feature values ​​arranged in a fixed order at the same time sampling point. By arranging the state vectors corresponding to multiple time sampling points in chronological order, a package is constructed. The system generates a behavioral state evolution sequence containing multiple time points and sends the sequence to the trend judgment unit 62. Based on multiple sets of feature information in the behavioral state evolution sequence, the trend judgment unit 62 calculates the time change gradient, fluctuation amplitude, and persistence offset index of each set of feature information. The time change gradient refers to the ratio of the change in feature value between two adjacent time sampling points to the time interval. The fluctuation amplitude refers to the difference between the maximum and minimum values ​​of the feature value within a preset time window. The persistence offset index is an index calculated based on the trend of the average change of continuous time sampling points. The system judges whether there is a risk trend of abnormal increase in a single feature information or synergistic increase in multiple feature information. If a risk trend is judged to exist, the system generates an event recognition result at the output of the main control chip 6.

[0034] In one embodiment, the response module 7 includes a local alert unit 71 and a remote communication unit 72, wherein: The local alert unit 71 is installed in the cockpit and is connected to the output of the main control chip 6. When the event recognition result indicates that the driver is distracted or has a tendency to drive while fatigued, the local alert unit 71 is used to trigger the voice broadcast device and the warning light device to output voice prompt information and visual flashing signal. The remote communication unit 72 is connected to the output of the main control chip 6 and is used to send alarm information to a preset emergency contact via wireless communication when the event identification result indicates that the alcohol level is too high or there is a high-risk driving trend. The alarm information includes the event type, alcohol concentration value, vehicle geographical location information and video clips.

[0035] Specifically, the local alert unit 71 is located in the cockpit and receives event recognition results via a wired connection with the output of the main control chip 6. When the event recognition result indicates a trend of inattention or fatigue driving, it controls the voice broadcast device and the warning light device to work simultaneously. The voice broadcast device is a device that can convert preset text content into a voice signal and output it through a speaker. The warning light device is a warning device that can emit light signals according to a preset flashing frequency and brightness. When triggered, the voice broadcast device outputs voice prompts corresponding to the trend of inattention or fatigue driving, and the warning light device outputs visual flashing signals at the same time. The remote communication unit 72 is connected to the output of the main control chip 6. When the event recognition result indicates a trend of excessive alcohol or high risk, it controls the voice broadcast device and the warning light device to work simultaneously. When driving, an alarm message is sent to a preset emergency contact via wireless communication. Wireless communication refers to a communication method that transmits data based on cellular mobile communication networks or satellite communication networks. The alarm message includes the event type, alcohol concentration value, vehicle geographical location information, and video clips. The event type refers to a text or code-based identifier determined according to the risk trend category in the event identification results. The alcohol concentration value refers to the numerical part of the alcohol concentration information collected by the alcohol sensor 1 and processed by the main control chip 6. The vehicle geographical location information refers to the vehicle's position coordinates in the geographic coordinate system collected by the positioning module 5. The video clips refer to the video data files continuously collected by the image acquisition module 3 within a preset time window and encoded and compressed by the main control chip 6.

[0036] In one embodiment, such as Figure 2 As shown, an intelligent drunk driving detection method includes: S10: Collect the driver's alcohol concentration information through the alcohol sensor 1 and send the alcohol concentration information to the main control chip 6.

[0037] Specifically, the alcohol sensor 1 is fixedly installed near the driver's breathing area. It detects changes in the concentration of ethanol molecules in the air through an internal gas detection unit. This gas detection unit uses a semiconductor sensing element or an infrared spectral analysis element to adsorb, react with, or spectrally scan an air sample, converting the ethanol molecule concentration into a corresponding electrical signal. When the driver exhales through the alcohol sensor 1, the gas detection unit generates a raw electrical signal representing the alcohol content in real time. This signal is amplified and filtered by the signal conditioning circuit inside the alcohol sensor 1 to remove background noise and irrelevant gas interference signals. The processed signal is then converted into numerical alcohol concentration information, which is transmitted to the first input terminal of the main control chip 6 via a wired data interface. S20: The voice acquisition module 2 acquires the driver's voice signal, extracts voice feature information, and sends the voice feature information to the main control chip 6.

[0038] Specifically, the voice acquisition module 2 is fixedly installed in the area directly in front of the driver inside the vehicle's front compartment. The voice acquisition module 2 includes a pickup unit and a signal processing unit. The pickup unit refers to the conversion of the driver's voice signal into an analog electrical signal using an electret microphone array or a condenser microphone element. The signal processing unit is a circuit structure that performs pre-amplification, bandpass filtering, and noise suppression on the analog electrical signal to remove environmental noise, engine vibration, and wind noise interference. Based on the clear voice signal output by the signal processing unit, the voice acquisition module 2 calculates the speech rate parameters and pitch parameters of the voice signal through a feature extraction algorithm. The speech rate parameter refers to the number of syllables pronounced per unit time, and the pitch parameter refers to the change pattern of pitch over time during pronunciation. The voice acquisition module 2 combines the calculated speech rate parameters and pitch parameters into voice feature information according to a preset data structure and transmits it to the second input terminal of the main control chip 6 through a wired data interface.

[0039] S30: The image acquisition module 3 acquires the image information in front of the vehicle, analyzes the image information based on the lane recognition algorithm, obtains lane departure feature information, and sends the lane departure feature information to the main control chip 6.

[0040] Specifically, the image acquisition module 3 is fixedly installed in the center of the front of the vehicle. The image acquisition module 3 includes an imaging unit and an image processing unit. The imaging unit refers to the use of a CMOS or CCD image sensor to convert the visible light scene in front of the vehicle into digital image data. The image processing unit refers to the hardware or software structure that performs white balance adjustment, brightness equalization and distortion correction on the digital image data to ensure the image quality for subsequent analysis. After obtaining the processed image information, the image acquisition module 3 uses a lane recognition algorithm to analyze the image information. The lane recognition algorithm refers to the identification of the lane line position in the road through image feature extraction methods such as edge detection, Hough transform and color threshold segmentation, and calculates the lateral offset and offset direction by combining the relative distance and angle between the vehicle's center position and the lane line. The lateral offset and offset direction are combined to form lane departure feature information, which is transmitted to the third input terminal of the main control chip 6 through a wired data interface.

[0041] S40: The vehicle's acceleration information is obtained through the dynamic detection module 4, the acceleration information is processed to obtain dynamic feature information, which includes emergency braking behavior feature information, lateral swaying behavior feature information and collision behavior feature information, and the dynamic feature information is sent to the main control chip 6.

[0042] Specifically, the dynamic detection module 4 is fixedly installed near the center of gravity inside the vehicle. The dynamic detection module 4 includes an acceleration sensor and a signal processing unit. The acceleration sensor is a MEMS accelerometer that can measure the linear acceleration components of the vehicle in three mutually perpendicular directions. The signal processing unit is a hardware or software structure used to perform noise filtering, zero-point calibration, and gravity component separation on the acquired raw acceleration signal. By extracting features from the processed acceleration information in the time and frequency domains, events with instantaneous acceleration changes greater than the emergency braking threshold are identified as emergency braking behavior features. Events with lateral acceleration periodically fluctuating for a short period of time and exceeding the lateral sway threshold are identified as lateral sway behavior features. Events with sudden acceleration peaks in any direction and peak durations less than the collision duration threshold are identified as collision behavior features. The above three types of dynamic feature information are combined with the corresponding timestamps to form structured dynamic feature information, which is then transmitted to the fourth input terminal of the main control chip 6 through a wired data interface.

[0043] S50: Obtain the vehicle's geographical location information through the positioning module 5 and send the geographical location information to the main control chip 6.

[0044] Specifically, the positioning module 5 is fixedly installed inside the vehicle near the roof to ensure the stability of satellite signal reception. The positioning module 5 includes a satellite signal receiving unit and a position calculation unit. The satellite signal receiving unit is a radio frequency receiver capable of receiving signals from the Global Positioning System (GPS), Global Navigation Satellite System (GLONASS), or BeiDou Navigation Satellite System (BDS). The position calculation unit calculates the vehicle's longitude, latitude, and altitude in the Earth's coordinate system using trilateration based on navigation message information from multiple satellites. It then combines this with time information obtained from the real-time clock module to generate timestamped geographic location information. After the geographic location information is generated, it is transmitted to the fifth input terminal of the main control chip 6 according to a preset data frame format via the serial data transmission interface built into the positioning module 5. S60: The main control chip 6 constructs a behavioral state evolution sequence from alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information and geographical location information within a preset time window, performs trend analysis on the behavioral state evolution sequence, determines whether there is an increasing risk trend, and generates event recognition results.

[0045] Specifically, after receiving the alcohol concentration information output by the alcohol sensor 1, the voice feature information output by the voice acquisition module 2, the lane departure feature information output by the image acquisition module 3, the dynamic feature information output by the dynamic detection module 4, and the geographical location information output by the positioning module 5, the control chip 6 performs timestamp alignment processing on the above-mentioned multiple types of feature information in its internal memory according to the start and end times of a preset time window to ensure that corresponding time sampling point records are formed under the same time reference. The control chip 6 then aligns the alcohol concentration value, voice feature parameters, lane departure feature parameters, dynamic feature parameters, and location coordinates contained in each time sampling point according to... Arranged chronologically, a behavioral state evolution sequence containing multiple time nodes is formed. After the behavioral state evolution sequence is constructed, a trend analysis program is called in the main control chip 6 to perform numerical change calculations within a sliding window for each type of feature information in the behavioral state evolution sequence. The least squares method is used to fit the slope of the change trend and calculate the fluctuation amplitude. At the same time, the trend slope is compared with the preset risk trend judgment threshold. When the trend slope is positive and exceeds the risk trend judgment threshold, the time window is judged to have an increasing risk trend, and an event recognition result containing risk type, relevant feature information and time range is generated in the main control chip 6.

[0046] S70: Output the event recognition result to the response module 7, which then performs the corresponding response operation based on the level of the event recognition result.

[0047] Specifically, after generating the event identification result, the main control chip 6 sends the event identification result to the response module 7 via a wired data interface. The sent content is organized according to a preset data frame format and includes three fields: the level of the event identification result, the risk type, and the time range. The level of the event identification result refers to the classification number used to indicate the risk intensity. After receiving the event identification result, the response module 7 performs integrity verification and field parsing on the data frame. Based on the internally stored response mapping table, it retrieves the response strategy that matches the level of the event identification result. The response mapping table is a table structure in the storage medium that records the correspondence between the level and the response parameters in the form of key-value pairs. The response parameters include the triggering method parameter, the drive duration parameter, and the triggering frequency parameter. After completing the retrieval, the response module 7 generates a control command according to the response parameters and outputs a control signal through the output port. The control signal is used to drive the connected prompt interface or control interface to perform the response operation. The response operation determines the start method according to the triggering method parameter, the holding duration according to the drive duration parameter, and the number of repetitions according to the triggering frequency parameter. After the response operation is completed, the response module 7 records the execution time and execution status and sends the execution status back to the main control chip 6 through the wired data interface to complete one processing loop.

[0048] In one embodiment, such as Figure 3As shown, in step S30, the image information is analyzed based on the lane recognition algorithm to obtain lane departure feature information, including: S301: The image acquisition module 3 acquires a continuous sequence of image frames in front of the vehicle, and performs edge enhancement processing on each image frame to obtain an image frame with enhanced edges.

[0049] Specifically, the imaging unit in image acquisition module 3 acquires a continuous sequence of image frames in front of the vehicle. The imaging unit is an image acquisition component consisting of a lens assembly, a photosensitive sensor array, and a driving circuit. The lens assembly focuses the light from the scene in front of the vehicle and projects the light onto the photosensitive sensor array. The photosensitive sensor array converts the light signal into an electrical signal. The driving circuit controls the photosensitive sensor array to sample at a set sampling frequency and output digital image data. The acquired image data is framed sequentially to form a continuous sequence of image frames. A continuous sequence of image frames refers to a set of image frames extracted according to the same time reference with consistent inter-frame intervals. After obtaining the continuous sequence of image frames... Each image frame undergoes edge enhancement processing. Edge enhancement processing refers to an image preprocessing process that sequentially performs grayscale conversion, noise suppression, and sharpening operations on a single frame to improve edge strength. Grayscale conversion involves converting color pixels into single-channel grayscale values ​​using preset weighting coefficients. Noise suppression involves replacing the center pixel with the median pixel value within a local window using median filtering to reduce the impact of random noise. Sharpening operation involves convolving the image with the Laplacian operator and linearly superimposing the convolution result with the original grayscale image using preset weights to enhance areas with rapid grayscale changes. After the above processing, intensity normalization is performed on the single frame image to keep pixel values ​​within a uniform quantization range, and the resulting image is output as an edge-enhanced image frame.

[0050] S302: Use an edge detection algorithm to process the edge-enhanced image frame and extract the left lane line boundary and the right lane line boundary.

[0051] Specifically, when processing the edge-enhanced image frame using an edge detection algorithm, the edge-enhanced image frame is first input into the edge detection operation flow. The edge detection algorithm is an image analysis method that identifies regions with significant gray-level changes by calculating the gradient changes in the spatial location of image pixel gray levels. In this step, the Canny operator is used as the edge detection operator. The Canny operator processing includes four stages: Gaussian smoothing, gradient calculation, non-maximum suppression, and double thresholding. Gaussian smoothing refers to performing convolution operations on the image using a two-dimensional Gaussian convolution kernel to reduce the interference of high-frequency noise on gradient calculation. Gradient calculation involves using a Gaussian convolution kernel to perform convolution operations on the image. The EL operator convolves the image in the horizontal and vertical directions and calculates the gradient magnitude and gradient direction of each pixel. Non-maximum suppression compares the gradient magnitudes of adjacent pixels along the gradient direction and retains the pixels with local maxima to refine the edge lines. Double threshold connection marks pixels with gradient magnitudes higher than the high threshold as strong edges and pixels with gradient magnitudes between the high and low thresholds as weak edges and connects them to the strong edges, ultimately forming a continuous edge contour curve. After extracting the edge contour curve, a boundary separation method based on line fitting is used to identify the left lane boundary located in the left half of the image and the right lane boundary located in the right half of the image.

[0052] S303: Based on the left lane line boundary and the right lane line boundary, the position of the lane center line in each image frame is estimated by a multi-frame curve fitting method to obtain the lane center line coordinate sequence.

[0053] Specifically, when estimating the lane centerline position in each image frame using a multi-frame curve fitting method based on the left and right lane boundaries, the pixel coordinate sequence of the left lane boundary and the pixel coordinate sequence of the right lane boundary are first paired one-to-one according to the vertical pixel rows of the image frame. In each vertical pixel row, the arithmetic mean of the horizontal coordinates of the left and right lane boundary pixels is calculated. This arithmetic mean is used as the horizontal coordinate of the lane centerline position in that vertical pixel row, and combined with the vertical coordinates of that vertical pixel row to form the lane centerline pixel coordinates, thus forming a set of lane centerline pixel coordinates. The set of lane centerline pixel coordinates from multiple consecutive frames is then combined in chronological order, and the multi-frame curve fitting method is used to fit the set of lane centerline pixel coordinates. The multi-frame curve fitting method refers to the mathematical processing of fitting a quadratic polynomial curve in a two-dimensional pixel coordinate system using the least squares method. The functional form of this quadratic polynomial curve is: Where y represents the vertical pixel coordinate value, x represents the horizontal pixel coordinate value, a represents the quadratic coefficient used to control the curvature of the curve, b represents the linear coefficient used to control the inclination of the curve, and c represents the constant term used to control the vertical offset position of the curve in the coordinate system; the optimal values ​​of coefficients a, b, and c are solved by minimizing the squared error between the actual vertical coordinates of the lane centerline pixel coordinates and the predicted vertical coordinates of the polynomial curve. The horizontal pixel coordinates of the fitted curve in each vertical pixel row are used as the lane centerline position coordinates, and finally a lane centerline coordinate sequence arranged in chronological order is formed and output for subsequent calculation of lane departure feature information.

[0054] S304: Based on the vehicle contour information in the image frame after edge enhancement, extract the coordinates of the vehicle center point to obtain the vehicle center point coordinate sequence.

[0055] Specifically, when extracting the vehicle center point coordinates based on the vehicle contour information in the edge-enhanced image frames, the image pixel matrix is ​​first scanned in each edge-enhanced image frame using a vehicle contour detection algorithm. The vehicle contour information refers to the set of closed curves formed by the boundary pixels of the vehicle shape. This set of closed curves is extracted by detecting the gradient changes of pixel gray values ​​and combining them with connected component analysis. After obtaining the vehicle contour information, the horizontal pixel coordinates of all pixels in the contour information are arithmetically averaged to obtain the horizontal pixel coordinates of the vehicle center point. The vertical pixel coordinates of all pixels in the contour information are arithmetically averaged to obtain the vertical pixel coordinates of the vehicle center point. The horizontal and vertical pixel coordinates of the vehicle center point are combined to obtain the vehicle center point coordinates. The above extraction operation is repeated in multiple consecutive edge-enhanced image frames, and the vehicle center point coordinate sequence is obtained by arranging them in chronological order.

[0056] S305: Compare the vehicle center point coordinate sequence with the lane center line coordinate sequence in the corresponding image frame, calculate the lateral offset distance of the vehicle relative to the lane center line, and obtain the lateral offset distance sequence.

[0057] Specifically, when comparing the vehicle center point coordinate sequence with the lane center line coordinate sequence in the corresponding image frame, firstly, in the image frame corresponding to each time sampling point, the vehicle center point coordinates in the vehicle center point coordinate sequence and the lane center line coordinates in the lane center line coordinate sequence are read respectively. The vehicle center point coordinates refer to the arithmetic mean of the horizontal and vertical coordinate values ​​of the pixels in the vehicle contour information, and the lane center line coordinates refer to the arithmetic mean of the horizontal coordinate values ​​of the left lane boundary coordinates and the right lane boundary coordinates. Then, the difference between the horizontal pixel coordinate values ​​of the vehicle center point coordinates and the horizontal pixel coordinate values ​​of the lane center line coordinates is calculated to obtain the lateral offset distance of the vehicle at that time sampling point. The lateral offset distance refers to the difference in pixel position of the vehicle center point relative to the lane center line in the horizontal direction. The above calculation process is repeated in all time sampling points, and the obtained lateral offset distances are arranged in chronological order to form a lateral offset distance sequence.

[0058] S306: Determine whether the lateral offset distances corresponding to multiple consecutive image frames in the lateral offset distance sequence all exceed a preset offset threshold. If the lateral offset distances corresponding to multiple consecutive image frames all exceed the preset offset threshold, then combine the turn signal status information in the corresponding image frames to determine whether there is a lateral offset behavior when the turn signal is not turned on. When it is determined that there is a lateral offset behavior when the turn signal is not turned on, generate lane departure feature information.

[0059] Specifically, when determining whether the lateral offset distances in multiple consecutive image frames in the lateral offset distance sequence all exceed a preset offset threshold, a sliding window detection is first performed on the lateral offset distance sequence. The length of the sliding window is equal to the preset number of consecutive frames. The lateral offset distance refers to the pixel difference between the coordinates of the vehicle's center point and the coordinates of the lane's center line in the horizontal direction. The preset offset threshold is a critical pixel distance value calculated based on road width information and vehicle width information. Within each sliding window, the lateral offset distance at each time sampling point within the window is compared with the preset offset threshold. When all lateral offset distances within the sliding window are greater than the preset offset threshold, the turn signal status information corresponding to each image frame within the window is read. The turn signal status information refers to the record of the vehicle's turn signal illumination and flashing status. When the turn signal status information indicates that the turn signal is in an off state at all time sampling points, the lateral offset distance sequence within that time period is marked as lateral offset behavior under the condition of no turn signal activation, and lane departure feature information is generated based on this lateral offset behavior. The lane departure feature information refers to the set of feature parameters used to characterize the lateral offset behavior that occurs when the vehicle's turn signal is not activated.

[0060] In one embodiment, such as Figure 4As shown, in step S306, if the lateral offset distances corresponding to multiple consecutive image frames all exceed a preset offset threshold, then, based on the turn signal status information in the corresponding image frames, it is determined whether there is a lateral offset behavior when the turn signal is not activated, including: S3061: Perform interpolation smoothing on the vehicle center point coordinate sequence extracted from multiple consecutive image frames, calculate the position change trend of the vehicle center point coordinate sequence in the horizontal direction, and obtain the vehicle driving direction change characteristics.

[0061] Specifically, when performing interpolation smoothing on the vehicle center point coordinate sequence extracted from multiple consecutive image frames, the vehicle center point coordinates extracted from multiple consecutive image frames are first arranged in chronological order to form a vehicle center point coordinate sequence. The vehicle center point coordinates refer to the position coordinate values ​​of the geometric center of the vehicle contour in the image plane coordinate system. In the vehicle center point coordinate sequence, for the horizontal coordinate component, a cubic spline interpolation method is used to generate transition coordinate points between adjacent sampling points to improve the smoothness of the position change curve. After obtaining the interpolated coordinate sequence, the horizontal coordinate component is smoothed using a moving average filtering method. The window length of the moving average filtering method is determined based on the image frame rate and the vehicle's driving speed to reduce coordinate jumps caused by instantaneous detection errors. Subsequently, the first-order difference of the smoothed horizontal coordinate component in the time dimension is calculated to obtain the rate of change of the vehicle center point in the horizontal direction. Based on the positive and negative signs and the magnitude of the rate of change, the driving direction change characteristics of the vehicle in the continuous time period are determined. The driving direction change characteristics refer to the set of feature parameters used to characterize the lateral movement direction and change trend of the vehicle relative to the lane centerline.

[0062] S3062: Match the vehicle's direction of travel change features with the turn signal status information extracted from multiple consecutive image frames to determine whether there is a turn signal activation signal during the period of continuous lateral position shift.

[0063] Specifically, when matching the vehicle's direction change features with the turn signal status information extracted from multiple consecutive image frames, the timestamps of the vehicle's direction change features and the turn signal status information are first aligned on the time axis. The turn signal status information refers to the left and right turn signal illumination status indicators obtained by detecting changes in image region brightness or parsing vehicle bus signals in consecutive image frames acquired by the image acquisition module. After completing the timestamp alignment, the feature values ​​representing the lateral movement direction in the vehicle's direction change features are compared frame by frame with the turn signal activation signals for the corresponding time period in the turn signal status information. If the vehicle's direction change features exhibit a stable unidirectional lateral offset trend during the period of continuous lateral position offset, and the turn signal status information does not detect the corresponding turn signal activation signal during this period, then this period is marked as the lateral offset time period when the turn signal is not activated, and the information of the lateral offset time period is stored in the lane departure feature information for subsequent event recognition result generation.

[0064] S3063: If no turn signal is detected during the period of continuous lateral position deviation, and the vehicle's direction of travel changes in accordance with the pattern of continuous lateral movement, then it is determined that there is lateral deviation behavior without the turn signal being activated.

[0065] Specifically, when determining whether there is lateral displacement behavior without turn signals during the time period of continuous lateral position displacement, the time period of continuous lateral position displacement is first determined based on the comparison results between the vehicle center point coordinate sequence and the lane center line coordinate sequence. This time period refers to the time range corresponding to the lateral displacement distance exceeding a preset displacement threshold in multiple consecutive image frames. Then, combined with turn signal status information, which refers to the vehicle's left and right turn signal illumination status indicators obtained through image area brightness change detection or vehicle bus signal parsing, the time period is analyzed frame by frame to determine whether no turn signal activation signal consistent with the lateral displacement direction is detected throughout the entire time period. After confirming that no turn signal activation signal is detected in the turn signal status information, pattern matching is performed on the vehicle driving direction change characteristics. The vehicle driving direction change characteristics refer to the lateral movement feature parameters calculated from the continuous change trend of the vehicle center point coordinates in the horizontal direction. The similarity of these feature parameters is compared with a pre-established continuous lateral movement pattern template. If the similarity is higher than the preset matching threshold, the vehicle driving direction change characteristics are determined to conform to the continuous lateral movement pattern, and this time period is marked as having lateral displacement behavior without turn signals.

[0066] In one embodiment, in step S40, the acceleration information is processed to obtain dynamic feature information, which includes emergency braking behavior feature information, lateral swaying behavior feature information, and collision behavior feature information, including: S401: Extract the corresponding linear acceleration components in the forward and backward directions from the acceleration information at each time sampling point, and determine whether the linear acceleration components in the forward and backward directions are lower than the emergency braking judgment threshold. If the linear acceleration components in the forward and backward directions are lower than the emergency braking judgment threshold, then the changes corresponding to the linear acceleration components in the forward and backward directions are used as emergency braking behavior feature information.

[0067] Specifically, when extracting the corresponding linear acceleration components in the forward and backward directions from the acceleration information at each time sampling point, a vehicle coordinate system is first established in the dynamic detection module 4, and the sensor's three-axis acceleration measurement results are mapped to the vehicle coordinate system. The linear acceleration components in the forward and backward directions refer to the acceleration components along the vehicle's direction of travel, with positive values ​​representing acceleration and negative values ​​representing deceleration. Subsequently, gravity component separation is performed on the mapped three-axis accelerations to obtain linear acceleration data free from the influence of static gravity. Gravity component separation refers to a processing method that uses low-pass filtering to estimate the gravity vector and subtracts the vectors from the original acceleration measurements to obtain linear acceleration. After obtaining the linear acceleration, the linear acceleration components in the forward and backward directions are selected, and the instantaneous deceleration value is calculated at each time sampling point. The emergency braking judgment threshold is used to distinguish between normal deceleration and rapid deceleration. Preset numerical parameters are given in negative form. When determining whether the linear acceleration components in the forward and backward directions are lower than the emergency braking judgment threshold, a sliding window denoising method is used to suppress instantaneous spike interference. The magnitude relationship between the filtered linear acceleration components in the forward and backward directions and the emergency braking judgment threshold is compared at the sampling point in the center of the window. When the linear acceleration components in the forward and backward directions are lower than the emergency braking judgment threshold, the changes corresponding to the linear acceleration components in the forward and backward directions are extracted as emergency braking behavior feature information. The changes include two timestamps, the start time and end time of emergency braking, to identify the time range of the emergency braking process; the minimum value of the linear acceleration components in the forward and backward directions during the emergency braking process as the peak deceleration; the integral result of the linear acceleration components in the forward and backward directions over time during the emergency braking process as the velocity change; and the number of continuous sampling points for the rising and falling edges to describe the waveform shape.

[0068] S402: Extract the corresponding left and right lateral acceleration components from the acceleration information at multiple consecutive time sampling points, determine whether there is an alternation between positive and negative directions in the lateral acceleration components, and if the amplitude of the alternation exceeds the lateral swaying judgment threshold, then use the change of the lateral acceleration components as the lateral swaying behavior feature information.

[0069] Specifically, when extracting the left-right lateral acceleration components from multiple consecutive time sampling points, a vehicle coordinate system is first established in the dynamic detection module 4, and the sensor's triaxial acceleration measurement results are mapped to the vehicle coordinate system. The left-right lateral acceleration components refer to the acceleration components perpendicular to the vehicle's direction of travel and parallel to the road surface, with positive values ​​representing rightward movement and negative values ​​representing leftward movement. Subsequently, gravity component separation is performed on the mapped triaxial accelerations to obtain linear acceleration data free from the influence of static gravity. Gravity component separation involves estimating the gravity vector using a low-pass filter and subtracting the vectors from the original acceleration measurements to obtain the linear acceleration. The processing method involves selecting the left-right lateral acceleration components after obtaining the linear acceleration and recording their numerical changes at multiple consecutive time sampling points. The lateral sway judgment threshold is a preset acceleration amplitude parameter used to distinguish between normal slight lateral swaying and obvious lateral swaying. To determine whether there is an alternation between positive and negative directions in the left-right lateral acceleration components, the sign change points are first detected in the time series, and the acceleration amplitude between adjacent sign changes is counted. When the acceleration amplitude of each positive-negative direction switch exceeds the lateral sway judgment threshold, the changes in the left-right lateral acceleration components at multiple consecutive time sampling points are used as lateral sway behavior feature information.

[0070] S403: At each time sampling point, calculate the rate of change of the acceleration vector based on the acceleration information, and determine whether the rate of change of the acceleration vector exceeds the collision determination threshold. If it exceeds the collision determination threshold, the change of the rate of change of the acceleration vector corresponding to that time sampling point is used as the collision behavior feature information.

[0071] Specifically, when calculating the rate of change of the acceleration vector based on acceleration information at each time sampling point, the three-axis acceleration measurement values ​​are first acquired in the dynamic detection module 4 and mapped to the vehicle coordinate system. The vehicle coordinate system refers to a rectangular coordinate system established with the vehicle's forward direction as the front-back axis, the vehicle's lateral direction as the left-right axis, and the vehicle's vertical direction as the up-down axis. Then, the gravity component of the mapped three-axis acceleration is separated to obtain linear acceleration data without the influence of static gravity. The linear acceleration data includes linear acceleration components in the front-back direction, lateral acceleration components in the left-right direction, and linear acceleration components in the vertical direction. After obtaining the linear acceleration data, the linear acceleration components in the three directions are... Acceleration components are combined in vector form to obtain the acceleration vector. The rate of change of the acceleration vector refers to the rate of change of the magnitude of the acceleration vector per unit time. When calculating the rate of change of the acceleration vector, the difference between the magnitudes of the acceleration vectors between adjacent time sampling points is calculated and divided by the time sampling interval to obtain the instantaneous rate of change. The collision determination threshold is a preset rate of change threshold parameter used to distinguish between acceleration changes in normal driving and collision impacts. When determining whether the rate of change of the acceleration vector exceeds the collision determination threshold, the instantaneous rate of change is compared with the collision determination threshold. When the instantaneous rate of change is greater than the collision determination threshold, the change in the rate of change of the acceleration vector corresponding to that time sampling point is used as the collision behavior feature information.

[0072] In one embodiment, in step S60, the main control chip 6 constructs a behavioral state evolution sequence based on alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information, and geographical location information within a preset time window, including: S601: Within a preset time window, extract the alcohol concentration value from the alcohol concentration information, the speech rate and tone parameters from the speech feature information, the lateral offset distance sequence from the lane departure feature information, the emergency braking behavior feature information, the lateral swaying behavior feature information and the collision behavior feature information from the dynamic feature information, and the location coordinates from the geographical location information.

[0073] Specifically, when extracting alcohol concentration values ​​from alcohol concentration information, speech rate and tone parameters from speech feature information, lateral deviation distance sequences from lane departure feature information, emergency braking behavior features, lateral swaying behavior features, and collision behavior features from dynamic feature information, and location coordinates from geographic location information within a preset time window, the main control chip 6 first calls the time synchronization module to define the start and end times of the preset time window. The preset time window refers to the continuous time interval used for statistical analysis of various feature data during event recognition and analysis. The time synchronization module is a functional unit that can synchronize data from different sources according to a unified time reference. After determining the preset time window, the alcohol concentration value is parsed from the alcohol concentration information collected by the alcohol sensor 1. The alcohol concentration value refers to the concentration of alcohol components in a unit volume of exhaled gas. Speech rate and tone parameters are extracted from the speech feature information obtained by the speech acquisition module 2. The speech rate parameter refers to the number of syllables uttered per unit time. The intonation parameter refers to the variation pattern of the voice signal in frequency and intensity. The lateral offset distance sequence is read from the lane departure feature information. The lateral offset distance sequence is a numerical sequence formed by arranging the lateral distance between the vehicle center point coordinates and the lane center line coordinates in consecutive image frames according to time. The emergency braking behavior feature information, lateral swaying behavior feature information, and collision behavior feature information are read from the dynamic feature information respectively. The emergency braking behavior feature information refers to the acceleration change recorded when the linear acceleration component in the front and rear directions is lower than the emergency braking judgment threshold. The lateral swaying behavior feature information refers to the acceleration change when the lateral acceleration component has alternating positive and negative changes and the amplitude exceeds the lateral swaying judgment threshold. The collision behavior feature information refers to the change when the rate of change of the acceleration vector exceeds the collision judgment threshold. The position coordinates are obtained from the geographical location information output by the positioning module 5. The position coordinates refer to the longitude and latitude values ​​of the vehicle in the geographic coordinate system. All the feature data extracted above are stored in the behavior state data buffer in chronological order for subsequent analysis.

[0074] S602: According to each time sampling point in the preset time window, the alcohol concentration value, speech rate parameter and tone parameter, lateral offset distance, emergency braking behavior feature information, lateral swaying behavior feature information, collision behavior feature information and position coordinates are synchronously aligned to construct a time series state vector set.

[0075] Specifically, when synchronizing and aligning the alcohol concentration value, speech rate and tone parameters, lateral offset distance, sudden braking behavior characteristics, lateral swaying behavior characteristics, collision behavior characteristics, and position coordinates at each time sampling point within a preset time window, the main control chip 6 first divides the time sampling point sequence according to the start and end times of the preset time window at fixed sampling intervals. A time sampling point refers to a discrete moment used to index the values ​​of various feature information under a unified time reference. Subsequently, the data records from the alcohol sensor 1, voice acquisition module 2, image acquisition module 3, dynamic detection module 4, and positioning module 5 are matched according to their timestamps, and the feature information records near each time sampling point are associated with that time sampling point. Synchronization and alignment means that at each time sampling point, the alcohol concentration value, speech rate parameter, and tone parameter are synchronized and aligned. The system selects the most recent valid records for each time sampling point, including tone parameters, lateral offset distance, emergency braking behavior characteristics, lateral swaying behavior characteristics, collision behavior characteristics, and position coordinates, and associates them one-to-one. If no valid record with a sufficiently close timestamp exists, the valid record from the previous time sampling point is retained to form a continuous value sequence. After synchronization and alignment, the alcohol concentration value, speech rate parameter, tone parameter, lateral offset distance, emergency braking behavior characteristics, lateral swaying behavior characteristics, collision behavior characteristics, and position coordinates are arranged in a fixed field order to form a state vector. The state vector refers to an ordered group of values ​​composed of the above-mentioned feature information at the same time sampling point. Finally, the state vectors corresponding to all time sampling points are combined in chronological order to form a time series state vector set.

[0076] S603: Perform behavioral modeling processing on the time series state vector set, extract behavioral change features, and generate a behavioral state evolution sequence representing the trend of driving behavior changes.

[0077] Specifically, when performing behavioral modeling processing on the time-series state vector set, the main control chip 6 first reads the state vectors in the time-series state vector set one by one in chronological order. The state vector is composed of alcohol concentration value, speech rate parameter, tone parameter, lateral offset distance, emergency braking behavior feature information, lateral swaying behavior feature information, collision behavior feature information, and position coordinates in a fixed field order. To eliminate the influence of numerical inconsistencies caused by different units, normalization processing is performed on the alcohol concentration value, speech rate parameter, tone parameter, and lateral offset distance. Normalization processing refers to normalizing within a preset time window. A linear transformation is performed on the values ​​centered at the mean and measured by the standard deviation to obtain a dimensionless sequence with zero mean and unit variance. After normalization, first-order difference sequences are calculated for alcohol concentration, speech rate, tone, and lateral offset. These first-order difference sequences represent the differences between adjacent time-series sampling points arranged chronologically, characterizing single-step changes. Simultaneously, the moving mean and moving standard deviation are calculated within a fixed-length sliding time window. The moving mean is the arithmetic average of the values ​​within the window, and the moving standard deviation represents the dispersion of the values ​​around the moving mean. This process is applied to sudden braking behavior. Event indication vectors are constructed from feature information, lateral swaying behavior feature information, and collision behavior feature information. An event indication vector is a binary sequence where the value is one when an event occurs at the corresponding time sampling point and zero when no event occurs. Within the same sliding time window, the number of sampling points with positive first-order differences in alcohol concentration, negative first-order differences in speech rate, positive first-order differences in lateral offset distance, and a single event indication vector are counted. The number of times these quantities occur simultaneously is also calculated. The number of times multiple conditions are met simultaneously within the same time sampling point is used to characterize the co-occurrence of multiple features. The normalized current value, the corresponding first-order difference, the moving average, the moving standard deviation, the event indication vector, and the number of times simultaneous occurrence are combined in a fixed field order to form behavior change features. Behavior change features are a set of feature parameters used to describe the relationship between numerical trends, fluctuation intensity, and event co-occurrence within a preset time window. The behavior change features are arranged sequentially according to the time sampling point order and a one-to-one correspondence is established with the state vector corresponding to the same time sampling point. These are then connected chronologically to form a behavior state evolution sequence representing the trend of driving behavior changes.

[0078] In one embodiment, step S60, namely performing trend analysis on the behavioral state evolution sequence to determine whether there is an increasing risk trend and generating event identification results, includes: S604: Risk scoring is performed on the behavioral change characteristics at each time sampling point in the behavioral state evolution sequence to obtain a risk score time series.

[0079] Specifically, when performing risk scoring on the behavioral change characteristics at each time sampling point in the behavioral state evolution sequence, the behavioral change characteristics in the behavioral state evolution sequence are read sequentially in the main control chip 6 according to time order. The behavioral change characteristics are composed of the normalized current value, first-order difference value, moving average, moving standard deviation, event indicator vector, and the number of times synchronization occurs, arranged in a fixed field order. The risk score value for each time sampling point is calculated according to a preset risk scoring rule. The risk scoring rule is a numerical calculation method based on linear weighting combined with threshold correction and amplitude limiting. A scoring weight coefficient is configured for each behavioral change characteristic field. The scoring weight coefficient is a non-negative real constant used in the risk scoring rule to measure the contribution of different behavioral change characteristics to the risk score. The calculation steps involve multiplying the behavioral change characteristic vector and the scoring weight coefficient vector element-wise and then summing the results to obtain the base score. Finally, the base score is determined based on a threshold. The scoring is corrected. Threshold determination refers to the process where, when the moving average of the alcohol concentration exceeds the alcohol threshold, or when the first difference of the lateral offset distance is positive and the number of simultaneous occurrences is greater than zero, or when any of the sudden braking, lateral swaying, or collision behavior features in the event indicator vector takes a value of one, the corresponding threshold addition is added to the base score. The threshold addition is a fixed score increment that corresponds to the threshold condition. To avoid abnormally large values ​​affecting the overall score stability, the corrected score is subjected to amplitude limiting. Amplitude limiting means that the score result is cropped to the interval between the preset minimum and maximum scores to obtain the risk score value of the current time sampling point. The risk score values ​​corresponding to each time sampling point are arranged sequentially according to the time sampling point order to form a risk score time series. The risk score time series is an ordered sequence of risk scores that covers all time sampling points within the preset time window and is arranged in chronological order.

[0080] S605: Perform trend fitting processing within a sliding window on the scoring time series, calculate the slope of the score change and determine whether the slope of the score change exceeds the risk trend judgment threshold. If the slope of the score change is positive and exceeds the preset risk trend judgment threshold, then it is determined that there is an increasing risk trend.

[0081] Specifically, when performing trend fitting within a sliding window on the scoring time series, the main control chip 6 first sets the length of the sliding time window for the scoring time series and slides it point by point on the time axis with a step size of one. The sliding time window refers to a fixed-length subsequence composed of continuous time sampling points, used to locally calculate the changing trend of the scoring time series. Within each sliding time window, a linear model is constructed using the time sampling point index as the independent variable and the corresponding risk score value as the dependent variable. The least squares method is used to fit the linear model. A fitting process is performed, where s represents the risk score, t represents the time sampling point number, k represents the slope of the score change, and b represents the intercept parameter. The least squares method is used to solve for the values ​​of k and b by minimizing the sum of the squared errors between the predicted and actual scores at all time sampling points within the window. After the fitting is completed, the slope of the score change, k, is read and compared with the risk trend judgment threshold. The risk trend judgment threshold is a preset lower limit of the slope used to determine whether the score time series shows a significant upward trend within a local window. When the slope of the score change is positive and exceeds the risk trend judgment threshold, the corresponding sliding time window is judged to have an increasing risk trend.

[0082] S606: Under the condition that there is a trend of increasing risk, generate the corresponding event identification result by combining the abnormal feature information in the behavioral state evolution sequence.

[0083] Specifically, after determining that there is an increasing risk trend, the main control chip 6 locates the time range corresponding to the increasing risk trend and retrieves the behavioral change features and state vectors within this time range from the behavioral state evolution sequence. Abnormal feature information is then screened from the behavioral state evolution sequence. Abnormal feature information refers to the set of feature information that meets the abnormal judgment conditions within a preset time window. The abnormal judgment conditions include: alcohol concentration exceeding the alcohol threshold; speech rate or tone parameters exceeding the normal range; lateral deviation distance exceeding the deviation threshold and the presence of lateral deviation behavior without turn signal activation in the lane departure feature information; the presence of sudden braking behavior feature information; the presence of lateral swaying behavior feature information; and the presence of collision behavior feature information. After completing the abnormal feature information screening, events are generated according to preset event assembly rules. Event identification results refer to structured records used to describe the content of risk events, including three fields: event type, time range, and event identification result level. The event type is determined based on the category of abnormal feature information. For example, when the blood alcohol concentration value is greater than the blood alcohol threshold, it is set as an alcohol overdose event; when the lateral deviation distance is greater than the deviation threshold and there is lateral deviation behavior without the turn signal being turned on, it is set as a lane departure event; when collision behavior feature information is present, it is set as a collision event; when multiple abnormal feature information is present at the same time, it is set as a high-risk driving event. The time range is taken from the start time and end time corresponding to the risk increase trend. The event identification result level is calculated by segmenting and mapping the slope of the score change in the scoring time series within the time range, and is written into the event identification result along with the event type and time range and output.

[0084] In one embodiment, the abnormal feature information in step S606 includes: S6061: In the alcohol concentration information, if the alcohol concentration values ​​at multiple consecutive time sampling points exceed the preset alcohol threshold, the corresponding changes in alcohol concentration values ​​will be used as alcohol concentration anomaly information in the abnormal feature information.

[0085] Specifically, when determining whether the alcohol concentration values ​​at multiple consecutive time sampling points exceed a preset alcohol threshold, the main control chip 6 first reads the alcohol concentration values ​​within a preset time window in chronological order and forms a sequence of alcohol concentration values ​​arranged by time sampling points. The preset alcohol threshold is a fixed numerical parameter used to distinguish between normal and excessive alcohol levels. The number of consecutive time sampling points is limited by a continuous point counting parameter, which refers to the minimum number of time sampling points required to continuously meet the judgment condition. During the sliding detection process, the alcohol concentration value sequence is scanned point by point with time sampling points as the step size, and the current consecutive points are counted in each scan. For the threshold exceeding count, when the alcohol concentration value continuously exceeds the preset alcohol threshold and the continuous threshold exceeding count is greater than or equal to the continuous point counting parameter, the alcohol concentration value change within the corresponding time range is extracted as alcohol concentration anomaly information. The alcohol concentration value change information includes two timestamps, the threshold exceeding start time and threshold exceeding end time, to identify the time range of continuous threshold exceeding; two statistical quantities, the peak alcohol concentration and the average alcohol concentration within the time range, to characterize the intensity level; and two time series quantities, the threshold exceeding duration and the maximum continuous threshold exceeding count, to characterize the persistence. The alcohol concentration anomaly information is written into the anomaly feature information for subsequent generation of event identification results.

[0086] S6062: In speech feature information, if the speech rate parameter is higher or lower than the normal speech rate range, or the tone parameter fluctuates within a preset time, the speech rate parameter and tone parameter at the corresponding time sampling point will be regarded as speech behavior abnormal information in the abnormal feature information.

[0087] Specifically, when determining whether the speech rate parameter is higher or lower than the normal speech rate range and whether the pitch parameter fluctuates within a preset time in the speech feature information, the main control chip 6 reads the speech feature information in chronological order and forms a speech rate parameter time series and a pitch parameter time series respectively. The normal speech rate range refers to the interval composed of the lower limit and upper limit values ​​of the speech rate parameters collected under normal driving conditions. The preset time refers to the continuous time length used to detect pitch changes and is expressed by the number of time sampling points. First, the speech rate parameter time series is compared point by point with the upper and lower limits of the normal speech rate range. When the speech rate parameter is lower than the lower limit of the normal speech rate range or higher than the upper limit of the normal speech rate range, the corresponding time sampling point is recorded and the time is extracted. The speech rate and intonation parameters at the sampling points are used as candidates for speech behavior anomalies. Then, a sliding time window is constructed for the intonation parameter time series according to a preset time length, and the fluctuation amplitude of the intonation parameter is calculated within each sliding time window. The fluctuation amplitude refers to the difference between the maximum and minimum values ​​of the intonation parameter within the window. The intonation fluctuation threshold is a fixed numerical parameter used to distinguish between normal intonation fluctuations and abnormal intonation fluctuations. When the fluctuation amplitude is greater than the intonation fluctuation threshold, the time sampling point corresponding to the center of the window is recorded, and the speech rate and intonation parameters at that time sampling point are extracted as candidates for speech behavior anomalies. Finally, the candidates for speech behavior anomalies caused by speech rate parameter exceeding the limit and the candidates for speech behavior anomalies caused by intonation parameter fluctuations are merged in chronological order and written into the abnormal feature information as speech behavior anomalies.

[0088] S6063: In lane departure feature information, if the lateral offset distance in multiple consecutive image frames exceeds the offset threshold and there is no turn signal status information, then the lateral offset distance sequence and the turn signal status combination information will be used as lane departure anomaly information in the anomaly feature information.

[0089] Specifically, when determining that the lateral offset distance exceeds the offset threshold in multiple consecutive image frames and there is no turn signal status information in the lane departure feature information, the main control chip 6 first reads the lateral offset distance sequence and turn signal status information and aligns them according to the time sampling points. The lateral offset distance sequence is a numerical sequence obtained by arranging the pixel differences between the vehicle center point coordinates and the lane center line coordinates in the horizontal direction in chronological order. The offset threshold is a critical pixel distance value used to distinguish between minor swaying within the normal lane and obvious movement towards the lane boundary. The turn signal status information is a time sequence identifier of the left and right turn signal illumination and flashing states obtained by the image acquisition module 3 or the vehicle bus. The lateral offset distance sequence is then detected window by window on the time axis according to the sliding window length set by the continuous point counting parameter. The continuous point counting parameter is an integer parameter used to limit the minimum number of time sampling points that continuously meet the judgment conditions. The algorithm compares the lateral offset distance of each time sampling point within a sliding window with the offset threshold. When the lateral offset distance of all time sampling points within the window is greater than the offset threshold, it reads the turn signal status information within the same time range and checks point by point whether there is a turn signal on in either the left or right direction. If no turn signal is detected within the time range, the value segment of the lateral offset distance sequence within the time range is combined with the state segment of the turn signal status information within the time range in chronological order to form the turn signal status combination information. The turn signal status combination information is an ordered record sequence consisting of a timestamp, a left turn signal status identifier, and a right turn signal status identifier. Subsequently, the value segment of the lateral offset distance sequence within the time range and the turn signal status combination information are written together into the abnormal feature information and marked as lane departure abnormal information.

[0090] S6064: In the dynamic feature information, if emergency braking behavior feature information, lateral swaying behavior feature information, or collision behavior feature information is detected within a preset time window, then the emergency braking behavior feature information, lateral swaying behavior feature information, or collision behavior feature information shall be regarded as dynamic driving abnormal information in the abnormal feature information.

[0091] Specifically, when determining whether emergency braking, lateral swaying, or collision behavior characteristics are detected within a preset time window in the dynamic feature information, the main control chip 6 reads the dynamic feature information according to the start and end times of the preset time window and traverses the records in chronological order. The dynamic feature information is a structured record set extracted by the dynamic detection module 4 based on acceleration information. Each record in the structured record set contains a timestamp, a behavior type field, and a corresponding numerical field. The behavior type field takes the values ​​of emergency braking, lateral swaying, or collision behavior characteristics. The numerical field in the emergency braking behavior characteristic information includes two timestamps: the start time and the end time of emergency braking, as well as the peak deceleration, the velocity change, the number of rising edge sampling points, and the number of falling edge sampling points. In the lateral swaying behavior characteristic information, it includes two timestamps: the start time and the end time of swaying, as well as the lateral acceleration component in the left and right directions. The maximum and minimum values, swaying period, and number of swaying events are included in the collision behavior feature information, which includes the time of the peak change rate, the peak change rate of the acceleration vector, and the duration exceeding the collision judgment threshold. When traversing the dynamic feature information within a preset time window, if a record with the behavior type field of emergency braking behavior feature information exists, that record is directly written into the abnormal feature information as dynamic driving anomaly information. If a record with the behavior type field of lateral swaying behavior feature information exists, that record is directly written into the abnormal feature information as dynamic driving anomaly information. If a record with the behavior type field of collision behavior feature information exists, that record is directly written into the abnormal feature information as dynamic driving anomaly information. Dynamic driving anomaly information refers to the set of records that are clearly identified as emergency braking behavior feature information, lateral swaying behavior feature information, or collision behavior feature information by the dynamic feature information within the preset time window, and the original timestamp and numerical fields are kept unchanged for subsequent assembly of event recognition results.

[0092] S6065: In geographic location information, if the driving path corresponding to the location coordinates of multiple time sampling points contains dangerous area identifiers or high-risk road segment labels, then the geographic location information will be marked as high-risk area information in the abnormal feature information.

[0093] Specifically, when determining whether the driving path corresponding to the location coordinates of multiple time sampling points contains dangerous area markers or high-risk road segment labels in the geographic location information, the main control chip 6 first reads the location coordinates of multiple time sampling points in chronological order and connects adjacent location coordinates in pairs to form a driving path polyline. The driving path refers to a spatial trajectory composed of multiple location coordinates arranged in chronological order and connected by line segments. Dangerous area markers refer to spatial area markers recorded in the storage medium as a sequence of polygon vertex coordinates, used to indicate geographic areas that require special attention. High-risk road segment labels refer to road segment markers recorded in the storage medium as road line segment coordinates and text or code markings, used to mark road line segments that require special attention. After the driving path is constructed, the spatial relationship between the driving path polyline and the dangerous area markers is calculated. The spatial relationship calculation includes point detection within the polygon and line segment intersection detection with the polygon boundary. In-shape detection refers to determining whether the coordinates of each position on the driving path polyline are located within the region enclosed by the vertex coordinate sequence of the polygon. Line segment intersection detection refers to determining whether any line segment of the driving path polyline has a geometric intersection with any boundary line segment of the polygon. At the same time, line segment proximity detection is performed on the driving path polyline and the high-risk road segment label. Line segment proximity detection involves calculating the minimum distance between a line segment of the driving path polyline and a road segment labeled as high-risk and comparing it with a road segment proximity threshold. The road segment proximity threshold is a fixed distance threshold parameter used to determine whether the driving path passes through a high-risk road segment. When the point is detected as being inside the polygon, or the line segment intersects with the polygon boundary, or the minimum distance of the line segment proximity detection is less than the road segment proximity threshold, the geographical location information recorded within the corresponding time range is marked as high-risk area information, and the high-risk area information is written into the abnormal feature information for use in the generation of subsequent event recognition results.

[0094] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0095] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. An intelligent drunk driving detection system, characterized in that, The intelligent drunk driving monitoring system includes an alcohol sensor (1), a voice acquisition module (2), an image acquisition module (3), a dynamic detection module (4), a positioning module (5), a main control chip (6), and a response module (7), wherein: An alcohol sensor (1) is placed in the driver's breathing area. The output of the alcohol sensor (1) is used to collect the driver's alcohol concentration information and output the alcohol concentration information to the first input of the main control chip (6). The voice acquisition module (2) is located inside the front cabin of the vehicle. The output of the voice acquisition module (2) is used to acquire the driver's voice signal, extract voice feature information, and transmit the voice feature information to the second input of the main control chip (6). The image acquisition module (3) is located at the front of the vehicle. The image acquisition module (3) is used to acquire the image information in front of the vehicle and to identify the image information in front of the vehicle based on the lane recognition algorithm to obtain lane departure feature information. The lane departure feature information is then transmitted to the third input terminal of the main control chip (6). The dynamic detection module (4) is located inside the vehicle. The dynamic detection module (4) is used to acquire the acceleration information of the vehicle, process the acceleration information to obtain dynamic feature information, the dynamic feature information includes emergency braking behavior feature information, lateral swaying behavior feature information and collision behavior feature information, and transmit the dynamic feature information to the fourth input terminal of the main control chip (6). The positioning module (5) is located inside the vehicle. The positioning module (5) is used to obtain the geographical location information of the vehicle and transmit the geographical location information to the fifth input terminal of the main control chip (6). The main control chip (6) is used to construct alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information and geographical location information within a preset time window, obtain a behavioral state evolution sequence, perform trend analysis on the behavioral state evolution sequence, and generate event recognition results; The response module (7) is located inside the vehicle and is connected to the output of the main control chip (6). The response module (7) is used to perform response operations based on the event recognition results generated by the main control chip (6).

2. The intelligent drunk driving monitoring system according to claim 1, characterized in that, The main control chip (6) includes a feature sequence construction unit (61) and a trend judgment unit (62), wherein: The feature sequence construction unit (61) is used to perform time-series sampling of alcohol concentration information, voice feature information, lane departure feature information and dynamic feature information respectively within a preset time window, construct a behavior state evolution sequence containing multiple time nodes, and send the behavior state evolution sequence to the trend judgment unit (62). The trend judgment unit (62) is used to calculate the time change gradient, fluctuation amplitude or continuous offset index of each set of feature information based on multiple sets of feature information in the behavioral state evolution sequence, and to determine whether there is a risk trend of abnormal rise of a single feature information or synergistic rise of multiple feature information. If it is determined that there is a risk trend, the event recognition result is generated at the output end of the main control chip (6).

3. The intelligent drunk driving monitoring system according to claim 1, characterized in that, The response module (7) includes a local alert unit (71) and a remote communication unit (72), wherein: The local reminder unit (71) is located in the cockpit and is connected to the output of the main control chip (6). When the event recognition result is characterized as a distraction or fatigue driving trend, the local reminder unit (71) triggers the voice broadcast device and the warning light device to output voice prompt information and visual flashing signal. The remote communication unit (72) is connected to the output of the main control chip (6) and is used to send alarm information to the preset emergency contact via wireless communication when the event identification result indicates that the alcohol level is too high or there is a high risk of driving. The alarm information includes the event type, alcohol concentration value, vehicle geographical location information and video clips.

4. A smart method for detecting drunk driving, characterized in that, The intelligent drunk driving detection method includes: The alcohol concentration information of the driver is collected by the alcohol sensor (1) and sent to the main control chip (6); The voice signal of the driver is collected by the voice acquisition module (2), the voice feature information is extracted, and the voice feature information is sent to the main control chip (6); The image acquisition module (3) acquires the image information in front of the vehicle, analyzes the image information based on the lane recognition algorithm, obtains the lane deviation feature information, and sends the lane deviation feature information to the main control chip (6). The vehicle's acceleration information is obtained through the dynamic detection module (4), and the acceleration information is processed to obtain dynamic feature information. The dynamic feature information includes emergency braking behavior feature information, lateral swaying behavior feature information and collision behavior feature information, and the dynamic feature information is sent to the main control chip (6). The location information of the vehicle is obtained through the positioning module (5) and sent to the main control chip (6); The main control chip (6) constructs a behavioral state evolution sequence based on alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information and geographical location information within a preset time window, performs trend analysis on the behavioral state evolution sequence, determines whether there is a risk increase trend, and generates event recognition results. The event identification result is output to the response module (7), which then performs the corresponding response operation according to the level of the event identification result.

5. The intelligent drunk driving detection method according to claim 4, characterized in that, The image information is analyzed based on the lane recognition algorithm to obtain lane departure feature information, including: The image acquisition module (3) acquires a continuous sequence of image frames in front of the vehicle and performs edge enhancement processing on each image frame to obtain an image frame with enhanced edges. The edge detection algorithm is used to process the edge-enhanced image frame to extract the left lane line boundary and the right lane line boundary; Based on the left lane line boundary and the right lane line boundary, the position of the lane center line in each image frame is estimated by a multi-frame curve fitting method to obtain the lane center line coordinate sequence. Based on the vehicle contour information in the edge-enhanced image frame, the coordinates of the vehicle center point are extracted to obtain the vehicle center point coordinate sequence; The vehicle center point coordinate sequence is compared with the lane center line coordinate sequence in the corresponding image frame to calculate the lateral offset distance of the vehicle relative to the lane center line, thus obtaining the lateral offset distance sequence. In the lateral offset distance sequence, it is determined whether the lateral offset distances corresponding to multiple consecutive image frames all exceed a preset offset threshold. If the lateral offset distances corresponding to multiple consecutive image frames all exceed the preset offset threshold, the turn signal status information in the corresponding image frames is combined to determine whether there is a lateral offset behavior when the turn signal is not turned on. When it is determined that there is a lateral offset behavior when the turn signal is not turned on, lane departure feature information is generated.

6. The intelligent drunk driving detection method according to claim 5, characterized in that, If the lateral offset distances corresponding to multiple consecutive image frames all exceed a preset offset threshold, then, based on the turn signal status information in the corresponding image frames, it is determined whether there is lateral offset behavior when the turn signal is not activated, including: Interpolation smoothing is performed on the vehicle center point coordinate sequence extracted from multiple consecutive image frames, and the position change trend of the vehicle center point coordinate sequence in the horizontal direction is calculated to obtain the vehicle driving direction change characteristics. The vehicle's direction of travel change features are matched with turn signal status information extracted from multiple consecutive image frames to determine whether there is a turn signal activation signal during the period of continuous lateral position shift. If no turn signal is detected during the period of continuous lateral displacement, and the vehicle's direction of travel changes in accordance with the pattern of continuous lateral movement, then it is determined that there is lateral displacement behavior without the turn signal being activated.

7. The intelligent drunk driving detection method according to claim 4, characterized in that, Acceleration information is processed to obtain dynamic characteristic information, which includes emergency braking behavior characteristic information, lateral swaying behavior characteristic information, and collision behavior characteristic information, including: At each time sampling point, the linear acceleration components in the forward and backward directions are extracted from the acceleration information. It is then determined whether the linear acceleration components in the forward and backward directions are lower than the emergency braking judgment threshold. If the linear acceleration components in the forward and backward directions are lower than the emergency braking judgment threshold, the changes in the linear acceleration components in the forward and backward directions are used as emergency braking behavior feature information. Extract the corresponding lateral acceleration components in the left and right directions from the acceleration information at multiple consecutive time sampling points, determine whether there is an alternation between positive and negative directions in the lateral acceleration components, and if the amplitude of the alternation exceeds the lateral swaying judgment threshold, then the change of the lateral acceleration components is used as the lateral swaying behavior feature information. At each time sampling point, the rate of change of the acceleration vector is calculated based on the acceleration information. It is then determined whether the rate of change of the acceleration vector exceeds the collision determination threshold. If it exceeds the collision determination threshold, the change in the rate of change of the acceleration vector at that time sampling point is used as the collision behavior feature information.

8. The intelligent drunk driving detection method according to claim 4, characterized in that, The main control chip (6) constructs a behavioral state evolution sequence based on alcohol concentration information, voice feature information, lane departure feature information, dynamic feature information, and geographical location information within a preset time window, including: Within a preset time window, extract the alcohol concentration value from the alcohol concentration information, the speech rate and tone parameters from the speech feature information, the lateral offset distance sequence from the lane departure feature information, the emergency braking behavior feature information, the lateral swaying behavior feature information and the collision behavior feature information from the dynamic feature information, and the location coordinates from the geographic location information. According to each sampling point in the preset time window, the alcohol concentration value, speech rate parameter and tone parameter, lateral offset distance, emergency braking behavior feature information, lateral swaying behavior feature information, collision behavior feature information and position coordinates are synchronously aligned to construct a time series state vector set; Behavioral modeling is performed on the time series state vector set to extract behavioral change features and generate a behavioral state evolution sequence that represents the trend of driving behavior changes.

9. The intelligent drunk driving detection method according to claim 4, characterized in that, Perform trend analysis on the behavioral state evolution sequence to determine if there is an increasing risk trend, and generate event identification results, including: Risk scores are assigned to the behavioral change characteristics at each time sampling point in the behavioral state evolution sequence to obtain a risk score time series. The trend fitting process within a sliding window is performed on the scoring time series, the slope of the score change is calculated, and it is determined whether the slope of the score change exceeds the risk trend judgment threshold. If the slope of the score change is positive and exceeds the preset risk trend judgment threshold, then it is determined that there is an increasing risk trend. Given the established trend of increasing risk, the corresponding event identification results are generated by combining abnormal feature information in the behavioral state evolution sequence.

10. The intelligent drunk driving detection method according to claim 9, characterized in that, Abnormal feature information, including: In the alcohol concentration information, if the alcohol concentration values ​​at multiple consecutive time sampling points exceed the preset alcohol threshold, the corresponding changes in alcohol concentration values ​​will be used as alcohol concentration anomaly information in the abnormal feature information. In speech feature information, if the speech rate parameter is higher or lower than the normal speech rate range, or the tone parameter fluctuates within a preset time, the speech rate parameter and tone parameter at the corresponding time sampling point will be regarded as speech behavior abnormal information in the abnormal feature information. In lane departure feature information, if the lateral offset distance in multiple consecutive image frames exceeds the offset threshold and there is no turn signal status information, then the lateral offset distance sequence and the turn signal status combination information are used as lane departure anomaly information in the anomaly feature information. If emergency braking behavior feature information, lateral swaying behavior feature information, or collision behavior feature information is detected within a preset time window in the dynamic feature information, then the emergency braking behavior feature information, lateral swaying behavior feature information, or collision behavior feature information will be regarded as dynamic driving abnormal information in the abnormal feature information. In geographic location information, if the driving path corresponding to the location coordinates of multiple time sampling points contains dangerous area markers or high-risk road segment labels, then the geographic location information will be marked as high-risk area information in the abnormal feature information.

Images