Resource-constrained terminal-oriented driving behavior multi-dimensional evaluation method, device and equipment

Abstract

The embodiment of the invention provides a resource-constrained terminal-oriented driving behavior multi-dimensional evaluation method, device and equipment. The method is applied to the technical field of intelligent driving, and comprises the following steps: dividing a current driving scene according to preprocessed vehicle driving data to obtain a driving scene division result; cutting the environment image by adopting a corresponding cutting strategy according to a driving scene division result; inputting the clipped environment image into a target detection model for analysis and processing to obtain a target detection result; the driving behavior is evaluated according to the target detection result and the vehicle motion state data, and triggering fine driving behavior judgment is conducted according to the driving behavior evaluation result; and scoring the driving behaviors by using the constructed multi-dimensional scoring model according to the types and the times of triggering the tiny driving behaviors in the vehicle driving process, and generating a driving diagnosis report. According to the method, the accuracy and the real-time performance of driving behavior evaluation on the vehicle-mounted equipment with limited resources are improved.

Description

Technical Field

[0001] This application relates to the field of intelligent driving technology, and in particular to a method, apparatus and device for multidimensional evaluation of driving behavior for resource-constrained terminals. Background Technology

[0002] With the increase in the number of motor vehicles, not only do traffic violations cause accidents, but many driving habits that are "not violations but lack quality" or "defensive awareness" also contribute to accidents. These include subtle driving habits such as not braking when passing through areas with obstructed vision, driving alongside large vehicles for extended periods, whether drivers actively slow down when approaching crosswalks, whether they yield to pedestrians, whether they alternate or slow down to observe when merging at intersections, whether they maintain a safe following distance, and whether they engage in aggressive overtaking or lane changing. Although most of these behaviors do not constitute traffic violations, they can truly reflect driving quality, habit stability, and potential safety risks, and are also important contributing factors to traffic accidents.

[0003] Current driving behavior assessment technologies are mainly divided into: traditional usage-based insurance (UBI) technology based on onboard automatic diagnostic systems and global positioning systems, and perception technology based on high-computing-power autonomous driving chips.

[0004] Traditional UBI technology based on in-vehicle automatic diagnostic systems and GPS can only detect simple indicators such as sudden braking, rapid acceleration, and speeding, and is completely unable to perceive the surrounding environment (such as whether there are people or whether it is an intersection), leading to biased assessments. Perception technology based on high-performance autonomous driving chips can accurately identify the environment, but it relies on expensive GPUs and complex deep learning models (such as Transformer), resulting in high power consumption and cost, and cannot be widely adopted in ordinary consumer-grade terminals such as dashcams.

[0005] Therefore, how to achieve "semantic-level" understanding and multi-dimensional quantitative evaluation of the aforementioned complex and subtle driving behaviors on ordinary chips with limited resources, using monocular cameras and low-cost sensors, without changing the vehicle structure, has become an urgent problem to be solved. Summary of the Invention

[0006] To address the aforementioned technical problems, this application provides a method, apparatus, and device for multidimensional evaluation of driving behavior on resource-constrained terminals.

[0007] Firstly, this application provides a multi-dimensional evaluation method for driving behavior on resource-constrained terminals, the method comprising:

[0008] Acquire vehicle driving data and preprocess the acquired vehicle driving data, which includes: environmental images and vehicle speed.

[0009] The current driving scenario is divided based on the preprocessed vehicle driving data to obtain the driving scenario division results;

[0010] Based on the driving scene segmentation results, the environmental image is cropped using the corresponding cropping strategy.

[0011] The cropped environmental image is input into the target detection model for analysis and processing to obtain the target detection results;

[0012] The driving behavior is evaluated based on the target detection results and vehicle motion state data, and the triggering of subtle driving behaviors is determined based on the driving behavior evaluation results.

[0013] The constructed multidimensional scoring model is used to score driving behavior based on the type and frequency of subtle driving behaviors triggered during vehicle operation, and a driving diagnostic report is generated.

[0014] Optionally, the preprocessing of the acquired vehicle driving data includes:

[0015] The nearest neighbor interpolation method is used to hard synchronize vehicle speed data with the timestamps of the environmental image to obtain the synchronized vehicle speed.

[0016] Optionally, the step of segmenting the current driving scenario based on the preprocessed vehicle driving data to obtain the driving scenario segmentation result includes:

[0017] The synchronized vehicle speed is compared with a preset first speed threshold and a second speed threshold, wherein the first speed threshold is greater than the second speed threshold;

[0018] If the vehicle speed after synchronization is greater than the first speed threshold and the duration exceeds the preset time, the driving scenario will be determined as high-speed mode.

[0019] If the vehicle speed after synchronization is less than the second speed threshold, the driving scenario will be classified as low speed mode.

[0020] If the vehicle speed after synchronization is greater than or equal to the second speed threshold and less than or equal to the first speed threshold, the driving scenario will be classified as medium speed mode.

[0021] Optionally, the step of cropping the environmental image using a corresponding cropping strategy based on the driving scene segmentation result includes:

[0022] In high-speed mode, the central region of the environmental image is cropped according to a first preset ratio;

[0023] In low-speed mode, the lower half and both sides of the environmental image are cropped according to a second preset ratio.

[0024] In medium speed mode, the environment image is not cropped.

[0025] Optionally, the step of evaluating driving behavior based on target detection results and vehicle motion state data, and determining subtle driving behaviors based on the driving behavior evaluation results, includes:

[0026] The lateral safety distance maintenance is evaluated based on the vehicle's lateral offset and target detection results. Based on the lateral safety distance maintenance evaluation results, it is determined whether the vehicle has triggered the first type of minor driving behavior.

[0027] The pedestrian crossing warning response is evaluated based on the vehicle's longitudinal acceleration and target detection results. Based on the pedestrian crossing warning response evaluation results, it is determined whether the vehicle has triggered the second type of subtle driving behavior.

[0028] The evaluation of alternating traffic flow at intersections is conducted based on the vehicle's longitudinal acceleration and target detection results. Based on the evaluation results, it is determined whether the vehicle has triggered the third type of subtle driving behavior.

[0029] The start-up response delay is evaluated based on the vehicle's start-up response time and the target detection results. Based on the start-up response delay evaluation results, it is determined whether the vehicle has triggered the fourth type of subtle driving behavior.

[0030] The evaluation of running a yellow light is conducted based on the distance between the vehicle and the stop line and the current speed. The evaluation results are used to determine whether the vehicle has triggered the fifth category of minor driving behavior.

[0031] Based on the target detection results and the current vehicle speed, a risk assessment of parallel driving of large vehicles is conducted, and the results of the risk assessment of parallel driving of large vehicles are used to determine whether the vehicle has triggered the sixth type of subtle driving behavior.

[0032] Optionally, the construction of a multi-dimensional scoring model to score driving behavior based on the type and frequency of subtle driving behaviors triggered during vehicle operation includes:

[0033] Based on the number of times various subtle driving behaviors are triggered in each dimension of the multidimensional scoring model during vehicle operation, and the corresponding negative scores for each subtle driving behavior, the scores of driving behaviors in each dimension are determined. The multidimensional scoring model includes: predictive ability dimension, politeness dimension, focus dimension, safety dimension, and stability dimension.

[0034] The scores of driving behavior in each dimension are weighted and summed according to the weights of each dimension of the multidimensional scoring model to obtain the driving behavior score result.

[0035] Secondly, this application provides a multi-dimensional evaluation device for driving behavior of resource-constrained terminals, comprising:

[0036] The acquisition module is used to acquire vehicle driving data;

[0037] The first processing module is used to preprocess the acquired vehicle driving data, which includes: environmental images and vehicle speed.

[0038] The second processing module is used to divide the current driving scenario based on the preprocessed vehicle driving data and obtain the driving scenario division result.

[0039] The second processing module is also used to crop the environmental image using a corresponding cropping strategy based on the driving scene segmentation result;

[0040] The second processing module is also used to input the cropped environmental image into the target detection model for analysis and processing to obtain the target detection result;

[0041] The driving behavior reasoning module is used to evaluate driving behavior based on target detection results and vehicle motion state data, and to determine the triggering of subtle driving behaviors based on the driving behavior evaluation results.

[0042] The driving behavior scoring module uses a constructed multi-dimensional scoring model to score driving behavior based on the type and frequency of subtle driving behaviors triggered during vehicle operation, and generates a driving diagnostic report.

[0043] Thirdly, this application provides a multi-dimensional assessment device for driving behavior of resource-constrained terminals, comprising:

[0044] Receiver; transmitter; memory; processor;

[0045] The receiver is used to receive instructions and data, and the transmitter is used to send instructions and data.

[0046] The memory stores computer-executed instructions;

[0047] The processor executes computer execution instructions stored in the memory to implement the multidimensional evaluation method for driving behavior of resource-constrained terminals as described in the first aspect and various possible implementations of the first aspect above.

[0048] Fourthly, this application provides a computer storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the multi-dimensional evaluation method for driving behavior of resource-constrained terminals as described in the first aspect and various possible implementations of the first aspect above.

[0049] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the multi-dimensional evaluation method for driving behavior on resource-constrained terminals as described in the first aspect and various possible implementations thereof.

[0050] This application provides a method, apparatus, and device for multi-dimensional evaluation of driving behavior on resource-constrained terminals. The method acquires vehicle driving data and preprocesses it, including environmental images and vehicle speed. Based on the preprocessed driving data, the current driving scenario is segmented to obtain a driving scenario segmentation result. The environmental image is then cropped using a corresponding cropping strategy based on the driving scenario segmentation result. The cropped environmental image is input into a target detection model for analysis to obtain a target detection result. Driving behavior is evaluated based on the target detection result and vehicle motion state data, and the triggered subtle driving behaviors are determined based on the driving behavior evaluation result. A constructed multi-dimensional scoring model is used to score the driving behavior based on the type and frequency of triggered subtle driving behaviors during vehicle operation, and a driving diagnostic report is generated. This improves the accuracy and real-time performance of driving behavior evaluation on resource-constrained in-vehicle devices. Attached Figure Description

[0051] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0052] Figure 1 A flowchart illustrating the multi-dimensional evaluation method for driving behavior on resource-constrained terminals provided in this application embodiment;

[0053] Figure 2 A schematic diagram of the structure of a multi-dimensional evaluation device for driving behavior on resource-constrained terminals provided in an embodiment of this application;

[0054] Figure 3 This is a schematic diagram of the structure of a multi-dimensional evaluation device for driving behavior on resource-constrained terminals provided in an embodiment of this application.

[0055] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0057] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein.

[0058] In this application, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0059] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0060] Figure 1 This is a flowchart illustrating the multi-dimensional evaluation method for driving behavior on resource-constrained terminals provided in an embodiment of this application. Figure 1 As shown, this embodiment provides a multi-dimensional evaluation method for driving behavior on resource-constrained terminals, including:

[0061] S1: Acquire vehicle driving data and preprocess the acquired vehicle driving data.

[0062] The vehicle driving data includes environmental images and vehicle speed.

[0063] Specifically, the acquired vehicle driving data is preprocessed, including: using the nearest neighbor interpolation method to hard synchronize the vehicle speed data with the timestamp of the environmental image to obtain the synchronized vehicle speed.

[0064] For example, video streams and GPS information are first collected simultaneously from the vehicle terminal. By analyzing the video, GPS speed, vehicle speed curves, and vehicle acceleration, the nearest neighbor interpolation method is used to hard synchronize the vehicle speed data with the timestamps of the video frames. Since different sensors have different sampling frequencies, time alignment is performed first. Due to the lack of an inertial measurement unit, acceleration cannot be directly read; GPS can only provide low-frequency speed data and has a 1-2 second lag, making it unable to capture "point braking" or "rapid acceleration." To achieve data alignment, the following interpolation method is used for synchronization:

[0065]

[0066] in, The synchronized vehicle speed, For video frame timestamps, GPS speed timestamp This is the GPS speed. Interpolation synchronization ensures that each video frame has a synchronized version of the vehicle speed.

[0067] Furthermore, a virtual reconstruction of the longitudinal acceleration is performed, satisfying the following formula:

[0068]

[0069] in, for Vehicle speed at any given moment This represents the time interval between two adjacent video frames. Vehicle acceleration is synchronized for each video frame; this longitudinal acceleration is used to determine sudden braking or rapid acceleration (e.g., ...). (This refers to actions involving rapid acceleration or sudden braking.)

[0070] S2: Divide the current driving scenario based on the preprocessed vehicle driving data to obtain the driving scenario division result.

[0071] Specifically, the synchronized vehicle speed is compared with a preset first speed threshold and a preset second speed threshold, wherein the first speed threshold is greater than the second speed threshold; if the synchronized vehicle speed is greater than the first speed threshold and the duration exceeds a preset time, the driving scenario is determined to be a high-speed mode; if the synchronized vehicle speed is less than the second speed threshold, the driving scenario is determined to be a low-speed mode; if the synchronized vehicle speed is greater than or equal to the second speed threshold and less than or equal to the first speed threshold, the driving scenario is determined to be a medium-speed mode.

[0072] For example, high-speed mode: when It is triggered when the duration exceeds 10 seconds. At this time, the vehicle is mainly in lane keeping and following mode, and the focus of vision is on the distance.

[0073] Low speed mode: When This is triggered at a specific time. At this point, the road conditions are complex, with pedestrians, non-motorized vehicles, traffic lights, vehicles ahead, and frequent intersection interactions.

[0074] Medium speed mode: When Triggered at time.

[0075] in, This is the high-speed threshold (e.g., 60km / h). The low speed threshold is (e.g., 30 km / h).

[0076] This division helps to dynamically adjust the region of interest (ROI), change the weight of behavior recognition, and optimize the model inference branches. It ensures that the algorithm selects different perception priorities in different road scenarios, thereby achieving the best detection accuracy and inference performance with limited computing power.

[0077] S3: Based on the driving scene segmentation results, crop the environment image using the corresponding cropping strategy.

[0078] Specifically, in high-speed mode, the central region of the environmental image is cropped according to a first preset ratio; in low-speed mode, the lower half and the two side regions of the environmental image are cropped according to a second preset ratio; and in medium-speed mode, the environmental image is not cropped.

[0079] For example, in high-speed mode, far-field cropping involves cropping a rectangular region representing the central 1 / 3 of the image's width and 1 / 3 of its height. This region covers distant vehicles ahead and lane line disappearance points, ignoring irrelevant roadside objects, thereby improving the recall rate for detecting small, distant targets.

[0080] Medium speed mode: In medium speed mode, the image is not cropped.

[0081] Wide field-of-view cropping in low-speed mode: Crop the lower half of the image (half the height) and the left and right edge areas (expanding to full width). This strategy simulates a driver's "scanning" habit, focusing on covering sidewalks, pedestrians on the roadside, and vehicles merging from the side, while discarding useless sky areas at the top of the image.

[0082] S4: Input the cropped environment image into the target detection model for analysis and processing to obtain the target detection results.

[0083] For example, scaling the cropped image to 320. For 320 or lower resolution, input to a lightweight object detection network (such as an improved YOLO-Fastest or NanoDet) that has undergone channel pruning and INT8 quantization. The network output layer only needs to retain core categories such as vehicles (Car / Truck / Bus), pedestrians, cyclists, lane lines, traffic lights, and road markings (zebra crossings / stop lines).

[0084] In an optional embodiment, after detecting a target ahead (such as a vehicle or pedestrian), a geometrically constrained ranging method is employed to assess following distance or collision avoidance risk. A lightweight target detection network is used to obtain a target detection bounding box, and the center point of the lower edge of the detection box is extracted as the target's "grounding point," with its ordinate in the image coordinate system denoted as... Based on the ground plane assumption, the longitudinal distance between the vehicle and the target can be simplified as follows:

[0085]

[0086] in, This indicates the installation height of the camera's optical center relative to the ground, expressed in meters (m). This indicates the equivalent focal length of the camera in the vertical direction, measured in pixels. It represents the vertical coordinate of the optical center (principal point) of the image in the pixel coordinate system. It is usually located near the center of the image height and records the distance between the vehicle and the vehicle directly in front of it in real time to determine whether the distance is safe based on the driving scenario.

[0087] S5: Evaluate driving behavior based on target detection results and vehicle motion state data, and determine the triggering of subtle driving behaviors based on the driving behavior evaluation results.

[0088] Specifically, driving behavior is evaluated based on target detection results and vehicle motion state data, and subtle driving behaviors are triggered based on the evaluation results, including:

[0089] (1) Evaluate the lateral safety distance maintenance based on the vehicle's lateral offset and target detection results, and determine whether the vehicle has triggered the first type of minor driving behavior based on the lateral safety distance maintenance evaluation results.

[0090] An example scenario for evaluating lateral safety distance maintenance is: visual detection of a cyclist / pedestrian to the left / right front.

[0091] Continuously track the lateral position deviation of this vehicle within the lane. Using the lane segmentation mask or key points output by a lightweight object detection network, the lane markings on the left and right sides of the current lane are fitted using the least squares method (to obtain...). and Set the origin of the image coordinate system to the top left corner of the image. The axis is horizontal to the right. The axis is vertically downward.

[0092] The equation for the left lane line is denoted as:

[0093] The equation for the right lane line is denoted as:

[0094] To reduce the impact of perspective distortion at a distance, a fixed line in the lower half of the image is selected as the reference line. , usually take (That is, at 80% of the image height), this position corresponds to the road surface close to the vehicle (approximately 5-10 meters away), where lane lines are clear and minimally affected by curvature. In the baseline... At each location, calculate the x-coordinates of the left and right lane lines respectively. and :

[0095]

[0096]

[0097] Calculate the pixel coordinates of the lane centerline in that row of the image.

[0098] Obtain the pixel coordinates of the image centerline (i.e., the optical axis projection corresponding to the center of the vehicle's front):

[0099]

[0100] in, The horizontal eccentricity error during camera installation can be obtained through factory calibration or long-term statistical averaging; it can be set to 0 by default. This represents the image resolution width. The pixel-level horizontal offset is then:

[0101]

[0102] Using the "standard lane width" as a scale reference, pixel deviations are converted into physical distances. Assuming a standard urban road lane width... Approximately 3.5 meters. Calculate the lane pixel width at the reference row:

[0103]

[0104] Physical lateral position offset:

[0105]

[0106] like ,express If the center of the lane is to the right of the center of the image, it means the vehicle is veering to the left.

[0107] like This means the vehicle is veering to the right.

[0108] If during the overdue period The system displays vehicle deviation and determines that the driver "possesses lateral safety awareness." If... (Maintaining center position) is judged as "lacking awareness of lateral avoidance".

[0109] (2) Evaluate the pedestrian crossing warning response based on the vehicle's longitudinal acceleration and target detection results, and determine whether the vehicle has triggered the second type of minor driving behavior based on the pedestrian crossing warning response evaluation results.

[0110] For example, the scenario for evaluating the pedestrian crossing warning response is: visual detection of a "bus stop" on the roadside or a stationary vehicle (obstruction source) in an adjacent lane and a pedestrian crossing.

[0111] The detection network can identify the specific semantic category of the target, and the longitudinal acceleration can be checked before reaching the target. ,like If a deceleration trend is detected (if a pedestrian is detected ahead, the speed must be reduced until a stop is reached), then the speed must be reduced to a stop. If the condition is met, it is considered "good defensive awareness". (At a constant speed or accelerating), it is judged as "insufficient driving anticipation".

[0112] (3) Evaluate the alternating traffic flow at the intersection based on the vehicle's longitudinal acceleration and target detection results, and determine whether the vehicle has triggered the third type of subtle driving behavior based on the evaluation results.

[0113] For example, the scenario for evaluating the alternating traffic flow at an intersection is: a merging sign is detected, and a vehicle is attempting to merge from the side ahead.

[0114] In the aforementioned lightweight target detection network, the following specific category labels are added: confluence markers, such as "Y"-shaped or narrow road markers; ground confluence arrows, typically arrows slanted to the left or right; guide lines / zebra stripe areas;

[0115] Once the detection network outputs confidence scores for any of the above categories... Mark the current scene state as (Suspected confluence) and initiated subsequent geometric verification. To eliminate near-range noise and far-range ambiguity, the system selected two key longitudinal distance points. and (For example , Using the inverse function of monocular ranging, these two physical distances are converted into image ordinates. and :

[0116]

[0117] in, This represents the ordinate of the image's optical center (principal point) in the pixel coordinate system. It utilizes the fitted lane line equation. and Calculate the pixel width at these two depth points:

[0118]

[0119] Nearby pixel width is The width of the distant pixels is .

[0120] To eliminate the width variation caused by image perspective effect, the aforementioned pixel width is... Converted to normalized physical width Then at the distance The estimated physical width at that location is:

[0121]

[0122] in, This represents the equivalent focal length of the camera in the horizontal direction, measured in pixels. On a standard straight road or a road with uniform curvature, the physical width of the lane markings is constant, i.e.:

[0123]

[0124] Define geometric convergence rate :

[0125]

[0126] Normal road determination: It should be close to 0, but small errors are permissible. (like ).

[0127] Merging / Lane Reduction Decision: If Continuously exceeding the preset convergence threshold (For example If the value is 0, it indicates that the physical width of the lane in the distance is shrinking significantly.

[0128] Once it is confirmed that the lane is narrowing ( This requires determining whether the lane ends on the left or right. Calculate the rate of change of the lane slope: extract... and slope parameter and Calculate its in The slope of the tangent line at the point.

[0129] Left-side import determination: When Compared to a normal straight track, the slope is significantly increased, while When the change is not significant (i.e., the left lane is steeper), it is judged as "the left lane is reduced / merging".

[0130] Right-side import determination: When Compared to a normal straight track, the slope is significantly increased, while When the change is not significant (i.e., the right lane is steeper), it is judged as "the right lane is reduced / merging".

[0131] If a target vehicle is present on the merging side, the "alternating traffic principle" evaluation logic is activated. The relative longitudinal distance between this vehicle and the adjacent vehicle is calculated. If the front of the car next to it is in front And this vehicle (Acceleration display braking), judged as "yielding"; if (Accelerating) is judged as "rushing".

[0132] (4) Evaluate the start response delay based on the vehicle’s start response time and the target detection results, and determine whether the vehicle has triggered the fourth type of subtle driving behavior based on the start response delay evaluation results.

[0133] For example, the scenario for evaluating start-up response delay is: standing still or waiting for the red light to turn green.

[0134] To reduce computational cost, sparse optical flow is used instead of full-image optical flow. This is applied to each frame of the image. In this process, the FAST corner detection algorithm is used to select a certain number of corner points as the feature point set to be tracked. To reduce interference from stationary objects in the sky or along the roadside, feature point selection should focus on the lower half of the image (road surface) and the central region. It is the first Each feature point in The vertical axis of time. Using the Lucas-Kanade optical flow algorithm, the tracking... Feature points in the next frame The corresponding position in ,Right now Calculate the longitudinal component of the average displacement vector. :

[0135]

[0136] If continuous Frames (e.g.) )satisfy If so, the vehicle is determined to be in a state of absolute stillness, usually Set to a minimum value (e.g.) (Pixels), used to filter noise caused by slight camera shake. In standby mode, the system continuously monitors the following two events to determine when the driver should begin to respond. .

[0137] Event 1: Determining when a red light turns green ( (Traffic light recognition) continuously monitors the status of recognized traffic lights during the waiting period. ,when After a preset duration (e.g., 1 second) of continuous and stable red light status is detected, the timer enters the ready state, and the red light turns green. Defined as The moment when the light first switches from red to green and is consistently and stably recognized as such. To prevent misjudgments due to recognition errors (such as mistaking a bright billboard for a green light), the required confidence level for green light recognition is [not specified]. Must be continuous Frames (e.g.) () is higher than the preset threshold.

[0138] Event 2: Determining the starting moment of the vehicle in front ( (Target Motion Analysis) When there is a vehicle obstructing the traffic light in the stationary waiting mode, or when there is no traffic light (e.g., following another vehicle at an unsignalized intersection), the response start time is the moment the vehicle in front begins to move. It identifies the nearest target vehicle located directly in front of the current vehicle. It was confirmed to be in a stationary state (by calculating the relative optical flow at its grounding point, its value is...). ). Target ahead Within the detection frame, calculate the average longitudinal displacement component of the target ground point of the preceding vehicle in the image. During the standby period, continuous monitoring is performed. At the start Defined as First consecutive Frames (e.g.) Exceeding the starting threshold At that moment.

[0139]

[0140] in It can be set to 1 pixel.

[0141] By combining the occurrence times of the two events, the final response start time can be determined:

[0142]

[0143] That is, the time when the first of the two events occurs is taken as the reference time when the driver should begin to respond.

[0144] Set a starting threshold (For example (pixels), this threshold represents the background moving at a perceptible speed relative to the camera, in order to avoid misjudgments caused by one-off noise impacts. Precisely defined as the average longitudinal displacement First consecutive (for example ) Frames exceed the starting threshold That moment:

[0145]

[0146] Calculate the final response delay using two precisely determined time points. and will With the preset delay limit (e.g., 2.5s) are compared to determine whether the driver is exhibiting "distraction upon starting".

[0147] (5) Evaluate whether the vehicle has triggered the fifth category of minor driving behavior based on the distance between the vehicle and the stop line and the current speed.

[0148] For example, the scenario for evaluating running a yellow light is when the green light turns yellow.

[0149] Calculate the distance from the stop line Safe braking distance It consists of two parts: reaction distance and braking distance ,Right now Reaction distance It refers to the distance the vehicle travels at its current speed from the moment the driver senses the need to brake (e.g., a yellow light illuminates) to the moment they begin to press the brake pedal; reaction time. Use industry standards or experience values, for example but Braking distance This refers to the distance traveled from the initial application of maximum braking until the vehicle's speed drops to zero. Based on the formula for uniformly decelerated linear motion:

[0150]

[0151] in, (for example (If the road surface is wet and slippery, the value will be smaller.) This is an empirical value set based on safety principles, which can be obtained from the road surface friction coefficient and gravitational acceleration. The calculated minimum safe braking distance is then used. Distance from actual stop line Compare, if the following conditions are met:

[0152]

[0153] Then it can be brought to a safe stop, among which This indicates a reserved safety distance (e.g., 2 meters) to compensate for system measurement errors, road condition uncertainties, and braking system response delays. If both of the following conditions are met simultaneously when the yellow light is detected, it is considered "aggressive yellow light violation":

[0154] Condition 1 (Can be stopped): Current distance Meets the conditions for safe braking:

[0155]

[0156] Condition 2 (Acceleration Behavior): After the yellow light illuminates, the calculated longitudinal acceleration of this vehicle continuously exceeds the acceleration threshold. (For example This logic allows for a precise distinction between forced behavior—"not having time to stop and having to rush through"—and aggressive behavior—"being able to stop but accelerating and rushing through"—and provides targeted driving evaluations.

[0157] (6) Based on the target detection results and the current vehicle speed, conduct a parallel risk assessment of large vehicles, and determine whether the vehicle has triggered the sixth type of subtle driving behavior based on the parallel risk assessment results of large vehicles.

[0158] For example, a scenario for assessing the risk of parallel driving with large vehicles is: driving alongside large vehicles at high speeds.

[0159] The target classification label belongs to truck, bus or large truck, and the target detection model outputs detection boxes. If the height of the detection box is in pixels Relative to width pixels ratio Significantly larger than ordinary cars (e.g.) This threshold can be calculated based on publicly available data.

[0160] By real-time detection of targets ahead in this lane longitudinal distance And calculate the minimum safe following distance. , ( Take an empirical value, such as a 2-second interval. If or The conditions for overtaking were deemed ripe, among which... This is an additional safe overtaking distance (e.g., 20 meters).

[0161] Furthermore, the length of this vehicle Take the empirical values ​​of car models, for example Meters. By continuously tracking the vertical pixel difference between the bottom edge (grounding point) and top edge of the target detection box in the image, and combining this with the aforementioned monocular ranging function, the length of a large target vehicle can be estimated. The geometric center of this vehicle is set as the reference point, i.e. Based on the aforementioned monocular ranging formula, calculate the longitudinal distance from the target's ground contact point to the vehicle's optical center. The target's center longitudinal position for:

[0162]

[0163] Calculate the longitudinal body boundary of this vehicle:

[0164]

[0165] The longitudinal boundary of the target vehicle body is If the farthest point of the boundaries of the two vehicles is greater than or equal to the nearest point of the two vehicles, it indicates that the two vehicles overlap longitudinally and are in parallel. Specifically, this is determined by judging the two longitudinal intervals. and The formula for determining whether there is an intersection is as follows:

[0166]

[0167] A non-empty state indicates that the two vehicles are driving side-by-side. To quantify the risk, the moment a parallel state begins, it is identified as a high-risk parallel state requiring the initiation of a timing function. Choosing to maintain parallel driving with a large vehicle on the side for an extended period, even when safe overtaking conditions are met, is considered a lack of safety awareness; the longer the duration, the greater the risk.

[0168] S6: Utilize the constructed multidimensional scoring model to score driving behavior based on the type and frequency of subtle driving behaviors triggered during vehicle operation, and generate a driving diagnostic report.

[0169] Specifically, based on the number of times various subtle driving behaviors are triggered in each dimension of the multidimensional scoring model during vehicle operation, and the corresponding negative scores for each subtle driving behavior, the scores of driving behavior in each dimension are determined. The multidimensional scoring model includes: predictive ability dimension, politeness dimension, attention dimension, safety dimension, and stability dimension. The scores of driving behavior in each dimension are weighted and summed according to the weights of each dimension of the multidimensional scoring model to obtain the driving behavior score result.

[0170] For example, a five-dimensional rating vector is constructed. These correspond to predictive ability, politeness, focus, safety, and stability, respectively. Each dimension starts with a score of 100. Then, scores are calculated based on the weights of each dimension. Receive the final total score For each trip, the number of times the aforementioned subtle behaviors are triggered is counted. Points will be deducted according to the following formula:

[0171]

[0172]

[0173] Among them, the specific driving behavior weight parameters ( The settings are as follows:

[0174] Anticipatory error: Ignoring the risks of pedestrians suddenly appearing on the street, crosswalks, and other pedestrians (10 points / time); running a yellow light (5 points / time).

[0175] Courtesy rating: Violation of alternating right-of-way (5 points / time), lack of lateral yielding (10 points / time).

[0176] Attention level: Initial response delay (2 minutes / time).

[0177] Safety: Large vehicles driving in blind spots alongside each other (5 points / time), following too closely (5 points / time).

[0178] Stability: Number of rapid accelerations / decelerations (3 minutes / time).

[0179] Regarding the scoring weights of the five dimensions The settings are: Safety (0.3), Predictability (0.25), Stability (0.2), Politeness (0.15), Focus (0.1).

[0180] In an optional embodiment, the generated rating report for a single trip includes a five-dimensional rating graph and a final score, and indexes the three timestamps with the highest deduction weights. The corresponding "high-risk moment" video clips (10 seconds before and after) and behavior tags are extracted from the storage card, and a "driving diagnostic report" is generated and pushed to the user's mobile phone.

[0181] This application provides a multi-dimensional driving behavior evaluation method for resource-constrained terminals. The method acquires vehicle driving data and preprocesses it, including environmental images and vehicle speed. The current driving scenario is segmented based on the preprocessed data to obtain a driving scenario segmentation result. The environmental images are then cropped using a corresponding cropping strategy based on the driving scenario segmentation result. The cropped environmental images are input into a target detection model for analysis to obtain target detection results. Driving behavior is evaluated based on the target detection results and vehicle motion state data, and minor driving behaviors are triggered based on the evaluation results. A constructed multi-dimensional scoring model is used to score the driving behavior based on the type and frequency of minor driving behaviors triggered during vehicle operation, and a driving diagnostic report is generated. This improves the accuracy and real-time performance of driving behavior evaluation on resource-constrained in-vehicle devices.

[0182] Figure 2 This is a schematic diagram of the structure of a multi-dimensional evaluation device for driving behavior on resource-constrained terminals provided in an embodiment of this application. Figure 2 As shown, the multi-dimensional driving behavior evaluation device 200 for resource-constrained terminals provided in this embodiment includes:

[0183] Module 201 is used to acquire vehicle driving data;

[0184] The first processing module 202 is used to preprocess the acquired vehicle driving data, which includes: environmental images and vehicle speed.

[0185] The second processing module 203 is used to divide the current driving scenario based on the preprocessed vehicle driving data to obtain the driving scenario division result;

[0186] The second processing module 203 is also used to crop the environmental image according to the driving scene segmentation result using a corresponding cropping strategy;

[0187] The second processing module 203 is also used to input the cropped environmental image into the target detection model for analysis and processing, and to obtain the target detection result;

[0188] The driving behavior reasoning module 204 is used to evaluate driving behavior based on target detection results and vehicle motion state data, and to determine the triggering of subtle driving behaviors based on the driving behavior evaluation results.

[0189] The driving behavior scoring module 205 uses a constructed multi-dimensional scoring model to score driving behavior based on the type and frequency of subtle driving behaviors triggered during vehicle operation, and generates a driving diagnostic report.

[0190] The multi-dimensional driving behavior evaluation device for resource-constrained terminals provided in this embodiment can execute the multi-dimensional driving behavior evaluation method for resource-constrained terminals provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0191] Figure 3 This is a schematic diagram of the structure of a multi-dimensional evaluation device for driving behavior on resource-constrained terminals provided in an embodiment of this application. Figure 3 As shown in the embodiment of this application, the multi-dimensional evaluation device for driving behavior of resource-constrained terminals 300 includes: a receiver 301, a transmitter 302, a processor 303, and a memory 304.

[0192] Receiver 301 is used to receive instructions and data;

[0193] Transmitter 302 is used to send commands and data;

[0194] Memory 304 is used to store computer-executed instructions;

[0195] The processor 303 is used to execute computer execution instructions stored in the memory 304 to implement the various steps of the multi-dimensional evaluation method for driving behavior on resource-constrained terminals in the above embodiments. For details, please refer to the relevant descriptions in the foregoing embodiments of the multi-dimensional evaluation method for driving behavior on resource-constrained terminals.

[0196] Optionally, the memory 304 can be either standalone or integrated with the processor 303.

[0197] When the memory 304 is set up independently, the electronic device also includes a bus for connecting the memory 304 and the processor 303.

[0198] This application embodiment also provides a computer storage medium storing computer execution instructions. When the processor executes the computer execution instructions, it implements the multi-dimensional evaluation method for driving behavior of resource-constrained terminals as described above by the multi-dimensional evaluation device for driving behavior of resource-constrained terminals.

[0199] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described multidimensional evaluation method for driving behavior on resource-constrained terminals.

[0200] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0201] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0202] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A multi-dimensional evaluation method for driving behavior on resource-constrained terminals, characterized in that, The method includes: Acquire vehicle driving data and preprocess the acquired vehicle driving data, which includes: environmental images and vehicle speed. The current driving scenario is divided based on the preprocessed vehicle driving data to obtain the driving scenario division results; Based on the driving scene segmentation results, the environmental image is cropped using the corresponding cropping strategy. The cropped environmental image is input into the target detection model for analysis and processing to obtain the target detection results; The driving behavior is evaluated based on the target detection results and vehicle motion state data, and the triggering of subtle driving behaviors is determined based on the driving behavior evaluation results. The constructed multidimensional scoring model is used to score driving behavior based on the type and frequency of subtle driving behaviors triggered during vehicle operation, and a driving diagnostic report is generated.

2. The method according to claim 1, characterized in that, The preprocessing of the acquired vehicle driving data includes: The nearest neighbor interpolation method is used to hard synchronize vehicle speed data with the timestamps of the environmental image to obtain the synchronized vehicle speed.

3. The method according to claim 2, characterized in that, The step of segmenting the current driving scenario based on the preprocessed vehicle driving data to obtain the driving scenario segmentation result includes: The synchronized vehicle speed is compared with a preset first speed threshold and a second speed threshold, wherein the first speed threshold is greater than the second speed threshold; If the vehicle speed after synchronization is greater than the first speed threshold and the duration exceeds the preset time, the driving scenario will be determined as high-speed mode. If the vehicle speed after synchronization is less than the second speed threshold, the driving scenario will be classified as low speed mode. If the vehicle speed after synchronization is greater than or equal to the second speed threshold and less than or equal to the first speed threshold, the driving scenario will be classified as medium speed mode.

4. The method according to claim 3, characterized in that, The step of cropping the environmental image according to the driving scene segmentation result using a corresponding cropping strategy includes: In high-speed mode, the central region of the environmental image is cropped according to a first preset ratio; In low-speed mode, the lower half and both sides of the environmental image are cropped according to a second preset ratio. In medium speed mode, the environment image is not cropped.

5. The method according to claim 1, characterized in that, The evaluation of driving behavior based on target detection results and vehicle motion state data, and the determination of subtle driving behaviors based on the evaluation results, include: The lateral safety distance maintenance is evaluated based on the vehicle's lateral offset and target detection results. Based on the lateral safety distance maintenance evaluation results, it is determined whether the vehicle has triggered the first type of minor driving behavior. The pedestrian crossing warning response is evaluated based on the vehicle's longitudinal acceleration and target detection results. Based on the pedestrian crossing warning response evaluation results, it is determined whether the vehicle has triggered the second type of subtle driving behavior. The evaluation of alternating traffic flow at intersections is conducted based on the vehicle's longitudinal acceleration and target detection results. Based on the evaluation results, it is determined whether the vehicle has triggered the third type of subtle driving behavior. The start-up response delay is evaluated based on the vehicle's start-up response time and the target detection results. Based on the start-up response delay evaluation results, it is determined whether the vehicle has triggered the fourth type of subtle driving behavior. The evaluation of running a yellow light is conducted based on the distance between the vehicle and the stop line and the current speed. The evaluation results are used to determine whether the vehicle has triggered the fifth category of minor driving behavior. Based on the target detection results and the current vehicle speed, a risk assessment of parallel driving of large vehicles is conducted, and the results of the risk assessment of parallel driving of large vehicles are used to determine whether the vehicle has triggered the sixth type of subtle driving behavior.

6. The method according to claim 1, characterized in that, The constructed multidimensional scoring model scores driving behavior based on the type and frequency of subtle driving behaviors triggered during vehicle operation, including: Based on the number of times various subtle driving behaviors are triggered in each dimension of the multidimensional scoring model during vehicle operation, and the corresponding negative scores for each subtle driving behavior, the scores of driving behaviors in each dimension are determined. The multidimensional scoring model includes: predictive ability dimension, politeness dimension, focus dimension, safety dimension, and stability dimension. The scores of driving behavior in each dimension are weighted and summed according to the weights of each dimension of the multidimensional scoring model to obtain the driving behavior score result.

7. A multi-dimensional evaluation device for driving behavior on resource-constrained terminals, characterized in that, The device includes: The acquisition module is used to acquire vehicle driving data; The first processing module is used to preprocess the acquired vehicle driving data, which includes: environmental images and vehicle speed. The second processing module is used to divide the current driving scenario based on the preprocessed vehicle driving data and obtain the driving scenario division result. The second processing module is also used to crop the environmental image using a corresponding cropping strategy based on the driving scene segmentation result; The second processing module is also used to input the cropped environmental image into the target detection model for analysis and processing to obtain the target detection result; The driving behavior reasoning module is used to evaluate driving behavior based on target detection results and vehicle motion state data, and to determine the triggering of subtle driving behaviors based on the driving behavior evaluation results. The driving behavior scoring module uses a constructed multi-dimensional scoring model to score driving behavior based on the type and frequency of subtle driving behaviors triggered during vehicle operation, and generates a driving diagnostic report.

8. A multi-dimensional assessment device for driving behavior on resource-constrained terminals, characterized in that, The device includes: Receiver; transmitter; memory; processor; The receiver is used to receive instructions and data, and the transmitter is used to send instructions and data. The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the multi-dimensional evaluation method for driving behavior of resource-constrained terminals as described in any one of claims 1-6.

9. A computer storage medium, characterized in that, The computer storage medium stores computer execution instructions, which, when executed by a processor, are used to implement the multi-dimensional evaluation method for driving behavior of resource-constrained terminals as described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, is used to implement the multidimensional evaluation method for driving behavior on a resource-constrained terminal as described in any one of claims 1-6.

Images