Automobile forward collision early warning and control system suitable for man-machine mixed driving traffic flow

Abstract

The invention discloses an automobile forward collision early warning and control system suitable for man-machine mixed driving traffic flow. The automobile forward collision early warning and control system comprises a decision-making center, a collision early warning system and an anti-collision control system. The collision early warning system collects driver information of a man-driven vehicle and traffic data of a current vehicle and a front vehicle in the driving process, and establishes different safe distance models according to the man-driven vehicle and the unmanned vehicle. Performing collision risk grade division and graded early warning on a result after data analysis according to a risk grade index, and sending a divided result to a decision center; the decision-making center judges whether to control the vehicle or not according to the collision risk level and sends a decision-making result to the anti-collision control system; performing vehicle speed or brake control on the unmanned vehicle according to the risk level; therefore, the driving safety of the mixed driving traffic flow is improved, and the occurrence of rear-end accidents is reduced.

Description

Technical Field

[0001] The present invention relates to the field of traffic technology, and in particular to an automobile forward collision warning and control system for a mixed human-machine driving traffic flow. Background Art

[0002] With the rapid development of the automobile industry and the continuous advancement of intelligent technology, the application of smart cars in road traffic has gradually become popular. In today's road environment, manually driven cars and smart cars coexist. This kind of human-machine mixed driving (abbreviated as "human-machine mixed driving") scenario is becoming increasingly complex, posing new challenges to traffic safety.

[0003] In mixed traffic flows where humans and machines are present, different vehicles have different levels of collision risks under the same circumstances, which places higher demands on existing collision warning and control systems in order to improve driving safety in mixed traffic flows and reduce the occurrence of rear-end collisions. Summary of the invention

[0004] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a vehicle forward collision warning and control system suitable for mixed human and machine traffic flow, thereby improving driving safety in mixed driving traffic and helping to reduce the occurrence of rear-end collisions.

[0005] To achieve the above object, the specific scheme of the present invention is as follows:

[0006] The first aspect of the present invention provides a vehicle forward collision warning and control system suitable for mixed human and machine driving traffic flow, including a decision center and a collision warning system and a collision avoidance control system;

[0007] The collision warning system collects information about the driver of a human-driven vehicle and traffic data about the current vehicle and the vehicle ahead during driving. It establishes different safety distance models for human-driven vehicles and unmanned vehicles respectively. It divides the collision risk level based on two different risk level indicators and issues graded warnings based on the data analysis results, and sends the divided results to the decision-making center.

[0008] The decision center determines whether to control the vehicle based on the collision risk level and sends the decision result to the collision avoidance control system;

[0009] The anti-collision control system controls the vehicle. When the human-driven vehicle fails to respond to the warning in time, that is, when the collision risk level becomes high, the automatic emergency braking system will be used to apply emergency braking to the vehicle, and the speed or brake control of the unmanned vehicle will be carried out according to the risk level.

[0010] Optionally, the collision warning system includes a data acquisition module, a data processing module and a graded warning module;

[0011] The data acquisition module is used to collect data on the driver information of the human-driven vehicle and the obstacles ahead, and obtain the driver information of the human-driven vehicle and the traffic data of the vehicle and the obstacles ahead; the data analysis module processes the traffic data of the vehicle and the obstacles ahead, and classifies the risks accordingly; the graded warning module issues different degrees of warnings based on different risk levels.

[0012] Optionally, the front obstacle data is data measured by solid-state laser radar and millimeter-wave radar respectively; the driver information of the human-driven vehicle includes driver age information, driving experience information, and driver fatigue level information.

[0013] Optionally, the data analysis module performs fusion processing on the obstacle data ahead, and the process of fuzzifying the driver information of the human-driven vehicle including the driver age information, the driving experience information, and the driver fatigue level information, and estimating the driver reaction time through fuzzy logic reasoning to obtain the collision risk level includes:

[0014] Extract features from the data to obtain features related to the safety distance model; through feature input, obtain a safety distance model for human-driven vehicles and a safety distance model for unmanned vehicles based on driver characteristics;

[0015] The fused data is compared with its corresponding safety distance, and the collision risk is graded using the collision risk index.

[0016] Optionally, the process of the hierarchical warning module dividing the risk levels of the classified data includes:

[0017] According to the classification of collision risk levels, warnings of corresponding degrees are given.

[0018] Optionally, the anti-collision control system includes a human-driven vehicle control module and an unmanned vehicle control module; when the human-driven vehicle fails to respond to the warning in time, that is, the collision risk level becomes high risk, the human-driven vehicle control module combines with the automatic emergency braking system to perform emergency braking on the vehicle, and the unmanned vehicle control module controls the vehicle speed or braking according to the collision risk level of the unmanned vehicle.

[0019] Optionally, if the risk level is level II risk or level III risk, the collision warning module performs a corresponding level II or level III warning;

[0020] Optionally, if the risk level is Level I risk, the human-driven vehicle control module will perform emergency braking in conjunction with the automatic emergency braking system upon receiving the control command from the decision center, and the unmanned vehicle control module will perform deceleration or braking control according to the command issued by the decision center.

[0021] The beneficial effects of the present invention are as follows: the present invention provides accurate information by collecting traffic data during vehicle driving and driver information of human-driven vehicles, thereby determining the risk level according to the risk level index and issuing an early warning, so as to help the driver of the human-driven vehicle to respond in time, or intervene in the vehicle, or effectively control the unmanned vehicle before a collision; thereby improving driving safety in a mixed driving environment and helping to reduce the occurrence of rear-end collisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the present invention;

[0023] Figure 2 It is a schematic diagram of the collision warning of the present invention;

[0024] Figure 3 It is a schematic diagram of the anti-collision control system of the present invention;

[0025] Figure 4 Flowchart of the classification method after data analysis module acquisition;

[0026] Figure 5 Schematic diagram of the classification method after data analysis module acquisition; DETAILED DESCRIPTION

[0027] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, but the implementation scope of the present invention is not limited thereto.

[0028] like Figures 1 to 3 As shown, the embodiment described in this invention is a vehicle forward collision warning and control system suitable for mixed human-machine driving traffic flow, including a decision center, a collision warning system and an anti-collision control system.

[0029] The collision warning system collects driver information of human-driven vehicles and traffic data of the current vehicle and the vehicle in front during driving, establishes different safety distance models according to human-driven vehicles and unmanned vehicles respectively, divides the results of data analysis into collision risk levels based on two different risk level indicators and issues graded warnings, and sends the divided results to the decision center; that is, the collision warning system divides the collision risk levels of the obstacles ahead and issues warnings.

[0030] The decision center determines whether to control the vehicle based on the collision risk level and sends the decision result to the collision avoidance control system;

[0031] The anti-collision control system exercises control over the vehicle. This control includes emergency braking of the vehicle in conjunction with the automatic emergency braking system when the human-driven vehicle fails to respond to the warning in time, that is, when the collision risk level becomes high risk, and braking control of the unmanned vehicle in conjunction with the automatic driving system.

[0032] This embodiment uses the collision warning system to collect traffic environment data and divide the risk levels during vehicle driving, and provides accurate information for the collision avoidance control system to help human-driven vehicle drivers respond in time, or intervene in the vehicle, or effectively control unmanned vehicles before collision; thereby improving driving safety in mixed driving environments and helping to reduce the occurrence of rear-end collisions.

[0033] The present embodiment describes a vehicle forward collision warning and control system applicable to mixed human-machine driving traffic flow. In some embodiments, the collision warning system includes a data acquisition module, a data processing module, and a graded warning module;

[0034] The data acquisition module is used to collect data on the driver information of human-driven vehicles and obstacles ahead, and obtain the driver information of human-driven vehicles and the traffic data of the vehicle and obstacles ahead; specifically, the data acquisition module uses the solid-state laser radar and millimeter-wave radar installed on the vehicle to perform comprehensive perception and data collection on the vehicle ahead in real time, and the driver information of human-driven vehicles is obtained through the on-board camera and the driver himself, and the collected data is sent to the data processing module; the data processing module processes the traffic environment data in real time to obtain the final risk level result; the graded warning module issues a warning of the corresponding level according to the classification of collision risk levels.

[0035] Specifically, traffic data includes the speed of the vehicle itself and the speed of the vehicle ahead. The driver information of the human-driven vehicle is extracted by using the on-board camera through a deep learning model to detect the driver's fatigue level. The driver's age information and driving experience information are manually input by the driver.

[0036] The vehicle forward collision warning and control system for mixed human-machine driving traffic flow described in this embodiment further includes: the data processing module processes the data to obtain the collision risk level, which includes:

[0037] Feature extraction is performed on the traffic data obtained by the data acquisition module. Deep learning technology and image segmentation technology can be used to extract key features that can reflect the distance to the vehicle in front, such as vehicle speed and vehicle distance.

[0038] This embodiment adopts multi-sensor data fusion and machine learning algorithms to more accurately calculate the distance to the obstacle ahead. Compared with the existing warning systems based on a single sensor or traditional algorithms, this embodiment has higher recognition accuracy and can better adapt to complex mixed driving environments.

[0039] The driver information of the human-driven vehicle obtained by the data acquisition module is used in the driver reaction time model based on fuzzy theory. The driver's age information, driving experience information and driver fatigue level that affect the driver's reaction time are used as the characteristic values ​​of the input model, and the characteristic values ​​are fuzzified. Through the logical operation of fuzzy rules, a fuzzy quantity of the driver's reaction time is obtained, and then defuzzification is used to convert this fuzzy quantity into an accurate and specific numerical value. This accurate numerical value is the driver's reaction time finally calculated by the model.

[0040] Three membership functions of driver reaction time, σ is 2:

[0041]

[0042] The weighted value is set to w, the weighted value of the driver's age is w1, the weighted value of the driver's driving experience is w2, and the weighted value of the fatigue level is w3. The sum of the three weighted values ​​is 1. Considering the influence of the three characteristic factors on the driver's reaction time, the change and influence of the fatigue level are relatively higher, and the influence of the driver's age and the driver's driving mileage on the result is similar, so the three weighted values, and are set to 0.3, 0.3 and 0.4 respectively. The membership of the three input characteristic values ​​is weighted according to the weighted value, and then a total membership value is output. According to the membership function of the output fuzzy variable, the accurate time value is calculated.

[0043] Furthermore, the process of data risk classification by the data processing module includes:

[0044] The driver's reaction time t is used as the input of the human-driven vehicle safety distance model:

[0045] Specifically, the vehicle in front is stationary: the driver's reaction time is t, the vehicle speed is v1, when the driver starts to decelerate, the maximum deceleration of the vehicle is a1, and the distance d0 maintained by the two vehicles when they are relatively stationary, then the safe distance D that the vehicle needs to maintain with the vehicle in front is:

[0046] D=v1t+v1 2 / 2a1+d0

[0047] The vehicle ahead has a constant speed: When the vehicle's own speed v1 is greater than the vehicle ahead's speed v2, the driver's reaction time is t, the vehicle's maximum deceleration is a1, and d0 represents the distance between the two vehicles when they are relatively stationary, then the safe distance D that the vehicle ahead needs to maintain is:

[0048] D=v1t+(v1 2 -v2 2 ) / 2a1-(v1v2-v2 2 ) / a1+d0

[0049] The vehicle ahead decelerates: The driver's reaction time is t. During this period, the vehicle itself remains in a uniform motion state with a speed of v1, and the vehicle ahead has a speed of v2. After that, the vehicle itself starts to brake and decelerate until the distance between the two vehicles is d0. At this time, the speed of the vehicle itself is less than or equal to the speed of the vehicle ahead, and the speeds are both v. The maximum decelerations of the two vehicles are a1 and a2 respectively. Then the safe distance D that the vehicle needs to maintain from the vehicle ahead is:

[0050]

[0051] Let d be the actual distance between the two vehicles and D be the safe distance. Compare d and D. Set the score to 1 when 0 < d - D <= 2m, set the score to 2 when -4m < d - D <= 0, and set the score to 3 when d - D <= -4m. Based on this, divide the collision risk levels:

[0052]

[0053]

[0054] Safe distance model for driverless vehicles: When the data of the obstacle ahead obtained by multiple sensors is less than the set value of the initial safe distance, the driverless vehicle will, according to the distance information and the current driving speed of the driverless vehicle, achieve the safe braking of the driverless vehicle by controlling the VCU control module of the driverless vehicle. Its safe distance model is as follows:

[0055] Let d br be the total safe distance for the driverless vehicle to brake, v be the driving speed of the driverless vehicle, v rel be the relative speed magnitude between the driverless vehicle and the obstacle ahead, a1 be the maximum deceleration that the driverless vehicle can achieve, a2 be the maximum deceleration that the target vehicle can achieve, t1 be the delay time of the intelligent vehicle brake, and d0 be the minimum braking distance size of the intelligent vehicle. Then:

[0056] d br = 0.5[v 2 / a1 - (v - v rel )^2 / a2] + e rel t1 + d0

[0057] The grading and warning module will give warnings corresponding to the levels according to the risk grading in the data processing module:

[0058]

[0059] The anti-collision warning unit performs corresponding control according to the collision risk level information transmitted by the decision-making center. When the collision risk level reaches Level III and the driver fails to respond in time, the human-driven vehicle will brake in conjunction with the automatic emergency braking system, while the unmanned vehicle will not issue a warning. When the risk reaches Level III, the automatic driving system will perform braking control.

[0060] According to different levels of collision risk, the collision warning system can achieve multi-level alarm response, so that the driver can clearly distinguish the urgency of the incident. Specifically, when the above-mentioned collision risk reaches the corresponding level, the graded warning module will issue an alarm of the corresponding degree, reminding the driver to take timely measures to avoid rear-end collisions and further ensure the safety of the driver; if the risk level is level II or level III, the graded warning module will activate the corresponding level of warning alarm; if the risk level is level I and the driver does not respond in time, the decision center sends information to the anti-collision control system, and the human-driven vehicle control module combines the automatic emergency braking system to perform braking control. Unmanned vehicles do not issue warnings. When receiving the level I risk sent by the decision center, the unmanned vehicle control module combines the automatic driving system to perform braking control.

[0061] Embodiment 2:

[0062] like Figure 3-5 ,The classification methods after data analysis module collection include:

[0063] Step S101, obtaining the driver information of the human-driven vehicle and the traffic data of the current vehicle and the vehicle ahead, wherein the driver information includes age, driving experience and fatigue level, and the traffic data includes vehicle speed and relative distance.

[0064] The driver's facial image is collected through the on-board camera, and the facial features are extracted using image processing technology. The driver's fatigue level is judged based on the facial features; the driver's age and driving experience information is obtained from the on-board information system, and the age and driving experience information are associated with the fatigue level information to obtain comprehensive driver status information; the vehicle-mounted sensor is used to collect the vehicle's speed information in real time, and the speed information broadcast by the vehicle in front is received through the on-board communication system to calculate the relative speed between the vehicle and the vehicle in front; the distance between the vehicle and the vehicle in front is measured using laser radar or millimeter-wave radar, and the relative position relationship between the vehicle and the vehicle in front is judged based on the relative speed information; the driver status information, vehicle speed information, relative speed information and relative position information are combined. The system integrates the information of different locations and constructs comprehensive traffic status data; based on the traffic status data, it uses a machine learning algorithm to evaluate the driving risk level and obtain the collision risk level in the current driving scenario; if the collision risk level exceeds the preset safety threshold, a collision warning is triggered, and different warning methods are used according to the risk level, such as voice prompts and instrument panel warnings; if the driver does not take timely measures after the warning, it is judged that the driver does not react in time, and the emergency braking function is automatically triggered to control the vehicle to slow down or stop to avoid a collision; it continuously tracks changes in the driver's status and traffic status, and updates the collision risk level in real time to ensure the timeliness and effectiveness of collision warning and control throughout the driving process.

[0065] The driver's facial image is collected by the on-board camera, and the facial features, including eyes and mouth, are extracted using the Haar feature classifier. The degree of fatigue is judged by the fuzzy comprehensive evaluation method based on parameters such as the proportion of eye closure time and yawning frequency. The degree of fatigue is divided into 5 levels, of which level 3 and above is fatigue driving. The driver's age and driving experience information are obtained from the on-board information system. The age and driving experience are quantified into empirical coefficients using the weighted average method, with a value range of 6-2, and weighted fused with the fatigue level information to form comprehensive driver status information. The on-board sensor collects the speed of the vehicle with a period of 100ms, receives the speed information of the vehicle in front through the DSRC communication protocol, and calculates the relative speed using the relative motion model with an accuracy better than 1m / s. The laser radar scans the 180° area in front with a period of 100ms to obtain the distance and angle information of obstacles. The millimeter-wave radar has a ranging accuracy of 0.2m, and the relative position with the vehicle in front is calculated by fusing the information of the two. The Kalman filter algorithm is used to fuse the driver's status, speed and relative position information to form traffic status data. The SVM algorithm is selected to evaluate the collision risk of traffic status data. It is trained through 500 sets of driving data and tested through 5 sets of typical scenarios, with an accuracy rate of over 95%. When the collision risk exceeds 6, a voice prompt warning is triggered. When it is higher than 8, the instrument panel flashes to warn. If the driver's reaction time exceeds 1s, the optimal deceleration curve is planned through a linear optimization algorithm to control the vehicle to slow down or stop to avoid a collision. The system updates traffic status data and collision risk levels in a cycle of 10ms, and dynamically adjusts the warning prompt time and braking intervention time according to the risk level to ensure the real-time and reliability of collision warning and active safety.

[0066] Image processing technology and fuzzy comprehensive evaluation method are used to extract the driver's facial features and evaluate the driver's fatigue level to obtain the driver's fatigue level information.

[0067] The on-board camera is used to collect the driver's facial image information in real time to obtain the driver's facial feature data. Based on the acquired driver's facial feature data, the driver's facial features are extracted through a deep learning model to obtain the key feature parameters reflecting the driver's fatigue state. The fuzzy comprehensive evaluation method is used to take the extracted driver's facial key feature parameters as evaluation indicators to establish a driver fatigue assessment model. According to the established driver fatigue assessment model, the extracted driver's facial key feature parameters are comprehensively evaluated to obtain the driver's fatigue score. The driver's fatigue score is divided into different fatigue levels according to the preset fatigue level threshold to determine the driver's current fatigue level. If the driver's fatigue level exceeds the preset safety threshold, a fatigue warning signal is triggered, and a fatigue prompt is issued through the on-board system to remind the driver to take a rest. If the driver's fatigue level is still not relieved after receiving the fatigue warning signal, the driving assistance system is started to properly control the speed and direction of the vehicle to ensure driving safety. According to the changes in the driver's fatigue level, the triggering threshold of the fatigue warning signal and the control strategy of the driving assistance system are dynamically adjusted to achieve adaptive monitoring and warning of the driver's fatigue state. The driver's fatigue monitoring data is uploaded to the cloud platform through the on-board communication system. Combined with the driver's historical fatigue data, a comprehensive assessment of the driver's fatigue driving risk is conducted to provide the driver with personalized fatigue driving prevention suggestions.

[0068] Exemplarily, the on-board camera collects the driver's facial image at a speed of 30 frames per second, locates the facial area through the face detection algorithm, and extracts the coordinate information of 68 key feature points on the face. The convolutional neural network model is used to process the facial feature data, and 10 key parameters reflecting the fatigue state, such as eye aspect ratio, blinking frequency, and mouth opening and closing, are extracted. The fuzzy comprehensive evaluation method is adopted, and these 10 parameters are used as evaluation indicators. By constructing a fuzzy matrix and calculating the fuzzy comprehensive evaluation vector, a comprehensive score of the driver's fatigue degree is obtained, and the score range is 0 to 100 points. According to expert knowledge and experience, the fatigue degree is divided into four levels: awake, mild fatigue, moderate fatigue, and severe fatigue, and the corresponding scoring thresholds are 20 points, 50 points, and 80 points respectively. When the driver's fatigue score exceeds 50 points, the system plays a voice prompt through the on-board audio, and displays a fatigue warning icon on the dashboard to remind the driver to take a rest. If the driver's fatigue score is still higher than 80 points for 5 consecutive minutes after receiving the warning, the driving assistance system will be activated to reduce the vehicle speed by 20% by controlling the electronic throttle and steering assistance, and assist the vehicle to stay in the lane. The system analyzes the fatigue monitoring data in the last hour every 10 minutes, and adaptively adjusts the fatigue warning threshold and auxiliary control strategy, such as raising the warning threshold when driving at high speed and reducing the auxiliary control strength when driving at night. Through the on-board 4G communication module, the fatigue monitoring data is uploaded to the cloud platform in real time. Using big data analysis technology, combined with the driver's fatigue status in the past month, the fatigue driving risk index is calculated, and personalized anti-fatigue driving suggestions are pushed to the driver according to the risk level, such as reasonable travel arrangements, proper rest, etc.

[0069] By fusing information from LiDAR and millimeter-wave radar, the Kalman filter algorithm is used to obtain the relative position with the vehicle in front and determine the distance and angle relationship with the vehicle in front.

[0070] Solid-state laser radar and millimeter-wave radar are used to scan the obstacles ahead to obtain information such as the distance and angle of the obstacles. The data obtained by the laser radar and millimeter-wave radar are registered in time and space, and the data of the two sensors are unified in the same coordinate system. According to the registered data, the point cloud segmentation algorithm is used to identify and classify the obstacles ahead to obtain information such as the type, position, and speed of the obstacles. The segmented obstacle information is input into the Kalman filter algorithm to estimate and predict the motion state of the obstacles to obtain a more accurate and smooth obstacle trajectory. According to the obstacle trajectory obtained by the Kalman filter, the relative position relationship between the vehicle and the vehicle in front is calculated, including the relative distance, relative angle, etc. If the relative distance is less than the safety threshold, it is judged that there is a collision risk and the collision warning process is entered; otherwise, the obstacle tracking is continued. In the collision warning process, according to the size of the relative distance, the collision risk level is divided into three levels: Level I (high risk), Level II (medium risk), and Level III (low risk). If the collision risk level is Level I, emergency braking is triggered immediately, and the vehicle is quickly decelerated until it stops by controlling the vehicle's braking system to avoid a collision. If the collision risk level is Level II or Level III, different warning methods will be adopted according to the risk level, such as sound warnings, flashing lights, etc., to remind the driver to pay attention to obstacles ahead and take braking, steering and other operations according to the situation.

[0071] For example, Velodyne VLP-16 solid-state laser radar and Bosch LRR4 millimeter-wave radar are used to scan the obstacles ahead. The laser radar has a scanning range of 360°, a scanning frequency of 10Hz, and a detection distance of up to 100m; the millimeter-wave radar has a detection distance of up to 250m and an angular resolution of 2°. The point cloud data obtained by the two sensors are spatiotemporally aligned, and the millimeter-wave radar point cloud is aligned to the laser radar coordinate system through the NDT algorithm, and the alignment error is controlled within 5cm. After alignment, the obstacle is segmented using a point cloud segmentation algorithm based on Euclidean clustering, and the obstacle is identified and classified in combination with the deep learning target detection algorithm YOLOv5, with an accuracy of 96%. The segmented obstacle information is input into the extended Kalman filter (EKF) algorithm, and the position and velocity of the obstacle are estimated and predicted by establishing a kinematic model of the obstacle, and the outliers are eliminated using a robust M estimator to obtain a smooth obstacle trajectory. According to the obstacle trajectory, the relative distance and angle between the vehicle and the vehicle in front are calculated. When the relative distance is less than the safety threshold (such as 50m), it is determined that there is a collision risk. According to TTC (collision time), the collision risk is divided into three levels: TTC less than 2s is level I (high risk), 2-5s is level II (medium risk), and 5-8s is level III (low risk). When the collision risk is level I, the vehicle ESC (electronic stability system) is controlled through the CAN bus to achieve a maximum deceleration of 8g, and the vehicle is decelerated by 50% within 1s to avoid a collision. When the collision risk is level II or level III, an audible and visual alarm is issued through the HMI system, and the size and flashing frequency of the warning icon of the HUD head-up display system are controlled according to the risk level. At the same time, pre-collision braking is performed through the HAP (semi-automatic parking) system to decelerate the vehicle by 20%.

[0072] According to the traffic status data, the SVM algorithm is used to evaluate the driving risk level and obtain the collision risk level in the current driving scenario.

[0073] The driver information of the human-driven vehicle and the traffic status data of the current vehicle and the vehicle in front during driving are obtained, including the speed of the vehicle, the speed of the vehicle in front, the distance between vehicles and other data. The acquired traffic status data is subjected to feature extraction, and the key features that can reflect the distance of the vehicle in front are extracted by using deep learning technology and image segmentation technology. According to the extracted features, the support vector machine (SVM) algorithm is used to evaluate the driving risk level to obtain the collision risk level in the current driving scenario. If the assessed collision risk level is level III risk, the collision warning system issues a warning with a small alarm sound; if the assessed collision risk level is level II risk, the collision warning system issues a warning with a medium alarm sound. If the assessed collision risk level is level I risk, the collision warning system issues a warning with a strong alarm sound, and sends the collision risk level information to the decision center. After receiving the level I risk information sent by the collision warning system, the decision center determines whether it is necessary to actively control the vehicle. If the decision center determines that it is necessary to actively control the vehicle, the control instruction is sent to the anti-collision control system. After the anti-collision control system receives the control command sent by the decision center, if the vehicle is human-driven and the driver does not respond in time, the anti-collision control system combines with the automatic emergency braking system to implement emergency braking on the vehicle; if it is an unmanned vehicle, the anti-collision control system combines with the automatic driving system to actively decelerate or brake the vehicle. Through the above steps, the driving risk assessment based on the SVM algorithm is realized, and according to the assessment results, the collision warning system, decision center, and anti-collision control system work together to implement graded warning and active safety control on the vehicle, thereby improving the driving safety in the mixed human-machine driving traffic flow.

[0074] Step S102, using fuzzy logic reasoning method to estimate the driver's reaction time according to the driver's age, driving experience and fatigue level.

[0075] The driver's age information, driving experience information and driver fatigue information are obtained as input feature values ​​of the fuzzy logic reasoning model; the input feature values ​​are fuzzified and converted into corresponding fuzzy quantities; according to the preset fuzzy rules, the fuzzy quantities are logically operated to obtain the fuzzy quantity of the driver's reaction time; the fuzzy quantity of the driver's reaction time is converted into an accurate specific value by the defuzzification method, which is the driver's reaction time finally estimated by the model; according to the estimated driver's reaction time, combined with traffic data such as the speed of the vehicle and the speed of the vehicle in front, the safe distance that the vehicle in front needs to maintain is calculated through the human-driven vehicle safety distance model; if the actual distance between the vehicle in front and the vehicle in front is less than the calculated safe distance, it is judged that there is a collision risk; according to the difference between the actual distance and the safe distance, the level of collision risk is determined through the preset scoring gradient; if the collision risk reaches level II or level III, the graded warning module issues an alarm of the corresponding degree to remind the driver to take measures; if the collision risk reaches level I and the driver does not respond in time, the human-driven vehicle control module of the collision avoidance control system combines with the automatic emergency braking system to implement braking control on the vehicle.

[0076] First, the system collects the driver's facial image through the on-board camera, analyzes the driver's facial features using the image processing algorithm, and estimates that the driver's age is 35 years old. At the same time, the system accesses the driver information database and obtains that the driver's driving experience is 8 years. In addition, by analyzing the driver's blinking frequency, yawning frequency and other physiological indicators, the fatigue detection model is used to judge that the driver's fatigue level is 6. The three characteristic values ​​of age, driving experience, and fatigue level are input into the fuzzy logic reasoning model. After the membership function conversion, the age is mapped to the "middle-aged" fuzzy set with a membership of 8; the driving experience is mapped to the "skilled" fuzzy set with a membership of 7; the fatigue level is mapped to the "fatigue" fuzzy set with a membership of 6. Then, the fuzzy inference engine performs fuzzy reasoning operations on these three fuzzy sets according to the preset fuzzy rule library, and obtains the driver's reaction time as a "slow" fuzzy set with a membership of 65. Finally, the "slow" fuzzy set is converted into a specific reaction time value through the centroid method of defuzzification, and the current reaction time of the driver is calculated to be 2 seconds. Next, the system obtains the real-time speed of the vehicle as 60km / h and the speed of the vehicle in front as 50km / h, and substitutes it into the safe distance model of the human-driven vehicle d=v^2 / (2*μ*g)+v*t+s0, where μ is the friction coefficient of 8, g is the acceleration of gravity, and s0 is the static safety distance of 2m, and the calculated safety distance is 41m. The actual distance between the vehicle and the vehicle in front is 38m measured by the millimeter-wave radar, which is less than the safe distance, and it is judged that there is a collision risk. Substitute the actual distance and the safe distance into the preset scoring gradient, 38<41, the difference is 3m, and the corresponding collision risk level is II. At this time, the graded warning module issues a yellow alarm and voice prompts the driver "Please maintain a safe distance, the distance to the vehicle in front is too close." If the actual distance is further reduced to 30m, the collision risk rises to level I and the system issues a red alarm. If the driver does not respond to braking within 5 seconds, the human-driven vehicle control module will take over the vehicle and control the automatic emergency braking system to decelerate until the distance between the vehicles returns to a safe range. Through the above method, the anti-collision control system can dynamically estimate the safe vehicle distance according to the driver's status and traffic conditions, provide timely warnings and active intervention, and effectively avoid the occurrence of collision accidents.

[0077] The driver's facial features are analyzed through image processing algorithms to obtain his age and other physiological indicators to determine the driver's fatigue level.

[0078] According to the driver's state input feature value, the fuzzy logic reasoning model is used for fuzzification processing to obtain the fuzzy set of the driver's reaction time.

[0079] Through millimeter-wave radar and safety distance model calculation, the difference between the actual distance and the safety distance of the vehicle is obtained to determine the collision risk level.

[0080] Step S103, establishing a safe distance model for human-driven vehicles according to the driver's reaction time and the vehicle's driving status.

[0081] The driver's age information, driving experience information and driver fatigue level are obtained as characteristic values ​​that affect the driver's reaction time; the characteristic values ​​are fuzzified using fuzzy theory, and the fuzzy value of the driver's reaction time is obtained through logical operations of fuzzy rules; the fuzzy value of the driver's reaction time is converted into an accurate and specific value using the defuzzification method as the driver's reaction time; a safe distance model for human-driven vehicles is established based on the vehicle's current speed, relative speed and distance to the vehicle in front; if the vehicle in front is stationary, the safe distance that the vehicle in front needs to maintain with the vehicle in front is calculated based on the driver's reaction time, the vehicle's speed, the vehicle's maximum deceleration and the distance that the two vehicles maintain when they are relatively stationary. Full distance; if the vehicle in front is traveling at a constant speed and the speed of this vehicle is greater than that of the vehicle in front, the safe distance that this vehicle needs to maintain with the vehicle in front is calculated based on the driver's reaction time and the maximum deceleration of the vehicle; if the vehicle in front is decelerating, the safe distance that this vehicle needs to maintain with the vehicle in front is calculated based on the driver's reaction time, the speed of this vehicle, the speed of the vehicle in front, and the maximum deceleration of the two vehicles; by comparing the actual distance with the calculated safe distance, the collision risk is graded according to the set scoring gradient; according to the graded collision risk level and the current driving status of the vehicle, the safe distance model of the human-driven vehicle is dynamically adjusted to achieve collision warning and auxiliary decision-making for the driver.

[0082] Exemplarily, firstly, the driver's age, driving experience and fatigue level are obtained through the driver information collection system. The age range is 18-60 years old, the driving experience range is 0-40 years, and the fatigue level range is 0-10. Then, the fuzzy theory is used to fuzzify these eigenvalues, for example, the age is divided into fuzzy sets such as "youth", "middle-aged", and "old", the driving experience is divided into fuzzy sets such as "novice", "skilled", and "veteran", and the fatigue level is divided into fuzzy sets such as "awake", "mild fatigue", and "severe fatigue". Then, through the logical operation of fuzzy rules, the various eigenvalues ​​are comprehensively analyzed to obtain the fuzzy amount of the driver's reaction time, such as "faster", "medium", and "slower". Then, the fuzzy amount of the reaction time is converted into an accurate value using the anti-fuzzification method such as the center of gravity method as the driver's reaction time, such as 8 seconds. After obtaining the driver's reaction time, a safe distance model for human-driven vehicles is established based on the parameters such as the speed of the vehicle, the relative speed and distance to the vehicle in front collected by the on-board sensor. When the vehicle ahead is stationary, use the formula d = vt + v 2 / 2a+d0 to calculate the safe distance, where v is the speed of the vehicle, t is the driver's reaction time, a is the maximum deceleration of the vehicle, and d0 is the safe distance when the two vehicles are relatively stationary, usually 2-5 meters. When the vehicle ahead is moving at a constant speed and its speed is less than the speed of the vehicle ahead, the formula d=vt+v is used. 2 / 2a calculates the safe distance. When the vehicle ahead slows down, use the formula d=v1t1+v1 2 / 2a1-v2 2 / 2a2 calculates the safe distance, where v1 and a1 are the speed and maximum deceleration of the vehicle, v2 and a2 are the speed and maximum deceleration of the front vehicle, and t1 is the driver's reaction time. By comparing the actual vehicle distance measured by the on-board millimeter-wave radar with the safe distance calculated by the model, the collision risk is graded according to the preset safety scoring standards and divided into three levels: "safe", "warning", and "dangerous". For example, when the actual vehicle distance is greater than 2 times the safe distance, it is "safe", when it is less than 8 times the safe distance, it is "dangerous", and when it is between the two, it is "warning". The system dynamically adjusts the parameters in the safe distance model based on the real-time assessment of the collision risk level, such as appropriately increasing the safe distance when driving at high speeds or in bad weather conditions. Finally, the warning information and auxiliary decision-making suggestions are promptly fed back to the driver to improve driving safety.

[0083] The reaction time is fuzzy processed according to the driver's age, driving experience and fatigue level to obtain the fuzzy value of the driver's reaction time.

[0084] By comparing the actual distance with the calculated safety distance, the collision risk level is judged and dynamically adjusted to provide the driver with early warning and auxiliary decision-making.

[0085] The defuzzification method is used to convert the fuzzy value of the driver's reaction time into an accurate value to determine the specific reaction time.

[0086] Step S104, using a multi-sensor fusion method to obtain the obstacle data ahead and establish a safe distance model for the unmanned vehicle.

[0087] Solid-state laser radar and millimeter-wave radar are used to collect data on obstacles ahead to obtain information such as the distance and speed of obstacles. The data collected by multiple sensors are fused through the Kalman filter algorithm to obtain the precise position and speed information of the obstacle ahead. According to the position and speed information of the obstacle, combined with the speed and braking performance parameters of the unmanned vehicle itself, a safe distance model for the unmanned vehicle is established. The safe distance threshold between the unmanned vehicle and the obstacle ahead is calculated in real time. If the actual distance between the unmanned vehicle and the obstacle ahead is less than the safe distance threshold, it is judged that there is a collision risk. According to the collision risk level, it is determined whether the speed or brake control of the unmanned vehicle is required. If control is required, a control instruction is sent to the VCU control unit of the vehicle through the unmanned vehicle control module. After receiving the control instruction, the VCU control unit controls the vehicle to perform the corresponding deceleration or braking operation until the distance between the vehicle and the obstacle ahead is greater than the safe distance threshold. The distance change between the vehicle and the obstacle ahead is continuously monitored, and the speed and braking control strategy of the vehicle is dynamically adjusted according to the calculation results of the safe distance model to ensure that the vehicle always maintains a safe distance from the obstacle ahead.

[0088] The unmanned vehicle uses Velodyne VLS-128 solid-state laser radar and Bosch LRR4 millimeter-wave radar to collect data on obstacles ahead. The detection distance of the laser radar can reach 200 meters, and the detection distance of the millimeter-wave radar can reach 250 meters. The scanning frequencies are 10Hz and 20Hz respectively. The obstacle position and speed data collected by the two sensors are fused through the Kalman filter algorithm. The obstacle position error after fusion is less than 1 meter, and the speed error is less than 0.5m / s. According to the position and speed information of the obstacle, combined with the unmanned vehicle's driving speed of 60 kilometers per hour and the maximum braking deceleration of 8g, a safety distance model is established: D = 0.039V^2+0.54V+5. The safety distance threshold at the current vehicle speed is calculated in real time. For example, the safety distance threshold at 60 kilometers per hour is 46 meters. If the actual distance is less than the safety distance threshold, it is judged that there is a collision risk and different control strategies are adopted according to the risk level: brake and decelerate immediately when the risk is high, slow down when the risk is medium, and send a warning signal when the risk is low. The unmanned driving control module sends control instructions to the vehicle's VCU control unit via the CAN bus. After receiving the instructions, the VCU control unit controls the vehicle to perform corresponding deceleration or braking operations, such as braking at a deceleration of 8g when the risk is high, until the distance between the vehicle and the obstacle in front is greater than the safe distance threshold. The system continuously monitors the distance changes between the vehicle and the obstacle at a cycle of 100ms, and adjusts the vehicle's speed and braking control strategy in real time to ensure that the vehicle always maintains a safe distance from the obstacle in front.

[0089] The data collected by multiple sensors are fused through the Kalman filter algorithm to obtain the precise position and speed information of the obstacle ahead.

[0090] The parameters of the driverless vehicle safety distance model are determined based on the obstacle position and the vehicle's own speed.

[0091] Calculate the distance change between the vehicle and the obstacle in real time, determine the collision risk level and adjust the control strategy.

[0092] Step S105, comparing the actual vehicle distance with the calculation result of the safety distance model to obtain a collision risk score.

[0093] Different safety distance models are established for manually driven vehicles and unmanned vehicles, among which the safety distance model for manually driven vehicles uses the driver's reaction time as one of the input parameters; multi-sensor data fusion technology is used to obtain the actual distance information between the vehicle and the obstacle in front, and key features such as vehicle speed and vehicle distance are extracted through deep learning and image segmentation technology; the extracted actual vehicle distance features are input into the corresponding safety distance model to obtain the calculation result of the safety distance model; the actual vehicle distance is compared with the calculation result of the safety distance model, and different scoring gradients are set to obtain the collision risk score; the collision risk level is determined according to the collision risk score, and 2m<actual vehicle distance-safety vehicle distance<5m is set to a score of 1 corresponding to level III risk, 0.5m<actual vehicle distance-safety vehicle distance<2m is set to a score of 2 corresponding to level II risk, and actual vehicle distance-safety vehicle distance<=0 is set to a score of 3 corresponding to level I level risk; judge the collision risk level, if it is level III risk, the graded warning module will issue a warning with a small alarm sound; if it is level II risk, the graded warning module will issue a warning with a medium alarm sound; if it is level I risk, the graded warning module will issue a warning with a strong alarm sound; if the collision risk level is level I risk and the driver of the human-driven vehicle does not respond in time, the decision center sends a control instruction to the human-driven vehicle control module, and the human-driven vehicle control module combines with the automatic emergency braking system to perform emergency braking on the vehicle; if the collision risk level is level I risk, the decision center sends a control instruction to the unmanned vehicle control module, and the unmanned vehicle control module combines with the automatic driving system to perform braking control on the vehicle; through the above steps, collision risk assessment based on safety distance model comparison is realized, and graded warning and active intervention control are carried out according to the assessment results, thereby improving driving safety in human-machine mixed driving traffic flow.

[0094] According to the different characteristics of manually driven vehicles and unmanned vehicles, safety distance models are established respectively. The safety distance model of manually driven vehicles takes the driver's reaction time as one of the input parameters. Assuming that the driver's reaction time is 5 seconds, the safety distance can be expressed as: safety distance = vehicle speed × 5 + vehicle length + margin. The safety distance model of unmanned vehicles takes the vehicle braking distance as the main parameter, safety distance = braking distance + vehicle length + margin. The multi-sensor data fusion technology is used to obtain the actual distance information between the vehicle and the obstacle in front. The Kalman filter algorithm is used to fuse the sensor data such as radar, lidar, and camera to obtain accurate vehicle distance information. At the same time, the convolutional neural network model in deep learning is used to segment the on-board camera image to extract key features such as vehicle speed and vehicle distance. The extracted actual vehicle distance features are input into the corresponding safety distance model to obtain the calculation results of the safety distance model. For example, when a manually driven vehicle travels at a speed of 60 kilometers per hour, the safety distance is 60×5+5+2=95 meters. The actual vehicle distance is compared with the calculation results of the safety distance model, and different scoring gradients are set to obtain the collision risk score. The collision risk level is determined based on the collision risk score. 2 meters < actual vehicle distance - safe vehicle distance < 5 meters is set to a score of 1, corresponding to level III risk; 0.5 meters < actual vehicle distance - safe vehicle distance < 2 meters is set to a score of 2, corresponding to level II risk; actual vehicle distance - safe vehicle distance ≤ 0 is set to a score of 3, corresponding to level I risk. Determine the collision risk level. If it is a level III risk, the graded warning module will issue a warning with a small alarm through the on-board speaker; if it is a level II risk, the graded warning module will issue a warning with a medium alarm; if it is a level I risk, the graded warning module will issue a warning with a strong alarm. If the collision risk level is a level I risk and the driver of the human-driven vehicle does not perform braking or other operations within 5 seconds, the decision center will send a control instruction to the human-driven vehicle control module. The human-driven vehicle control module will perform emergency braking on the vehicle in conjunction with the automatic emergency braking system, and the braking deceleration can reach 5 meters / second. 2 If the collision risk level is level I, the decision center sends a control command to the unmanned vehicle control module, and the unmanned vehicle control module combines the automatic driving system to implement braking control on the vehicle, with a braking deceleration of up to 8 meters per second. 2 Through the above steps, collision risk assessment based on safety distance model comparison is realized, and graded warning and active intervention control are carried out according to the assessment results, which improves driving safety in human-machine mixed driving traffic flow.

[0095] The multi-sensor data is fused through the Kalman filter algorithm to obtain accurate vehicle distance information.

[0096] Convolutional neural network is used to segment the vehicle camera images and extract key features such as vehicle speed and vehicle distance.

[0097] The actual vehicle distance characteristics obtained are input into the safety distance model to judge the collision risk level and determine the warning level.

[0098] Step S106, classifying the risk levels according to the collision risk score, including level I, level II, and level III risks.

[0099] The steps of classifying the collision risk level according to the collision risk index are as follows: construct a safe distance model for human-driven vehicles and a safe distance model for unmanned vehicles based on driver characteristics. By inputting the fused data into the model, the distance output is obtained. The collision risk level is determined according to the risk level index. If the risk level is level I risk, the human-driven vehicle control module performs emergency braking in combination with the automatic emergency braking system when receiving the control command from the decision center, and the unmanned vehicle control module performs deceleration or braking control according to the command issued by the decision center. If the risk level is level II risk or level III risk, the collision warning module performs corresponding level II or level III warning. According to the classification result of the collision risk level, the collision warning module performs a warning of the corresponding level. The decision center determines whether it is necessary to control the vehicle according to the collision risk level classification result transmitted by the collision warning module. If the decision center determines that it is necessary to control the vehicle, the control command is sent to the human-driven vehicle control module or the unmanned vehicle control module of the collision prevention control system. The human-driven vehicle control module or the unmanned vehicle control module of the collision avoidance control system takes corresponding vehicle control measures, such as emergency braking, deceleration, etc., according to the control instructions of the decision-making center to avoid collision accidents.

[0100] First, a safe distance model for human-driven vehicles based on driver characteristics is constructed, considering that the driver's reaction time is usually 2-8 seconds and the braking deceleration is 5-8m / s. 2 , establish the safety distance calculation formula D = v 2 / (2a)+vt. At the same time, a safe distance model for unmanned vehicles is constructed, considering that the vehicle's perception-decision-execution delay is within 3 seconds and the braking deceleration can reach 10m / s 2 , establish the safety distance calculation formula D = v 2 / (2a)+3v. The data collected by sensors such as millimeter-wave radar, laser radar, and ultrasonic radar are fused through Kalman filtering and input into the safety distance model to calculate the real-time safety distance. According to the comparison between the safety distance and the threshold, the collision risk level is divided. When the safety distance is less than 2 meters, it is a level I risk, 2-5 meters is a level II risk, and 5-10 meters is a level III risk. When it is judged as a level I risk, the human-driven vehicle control module receives the control command from the decision-making center and activates the automatic emergency braking system at a speed of 10m / s 2The unmanned vehicle control module receives the command from the decision center and brakes at a deceleration rate of 12m / s 2 The vehicle brakes at a deceleration rate of 100%. When the risk is judged to be level II or level III, the collision warning module warns with a buzzer frequency of 2Hz and 1Hz respectively, and displays yellow and orange warning information on the display screen. The decision center analyzes the collision risk level and traffic scene comprehensively, and sends a braking command to the human-driven vehicle control module or the unmanned vehicle control module when it is judged that the collision is unavoidable. The control module executes the instructions of the decision center, controls the vehicle to emergency brake or decelerate through the actuator, and starts the controllable deceleration mode to minimize the collision loss.

[0101] By fusing the data from sensors such as millimeter-wave radar, lidar, and ultrasonic radar through Kalman filtering, a more accurate real-time safety distance can be obtained.

[0102] Step S107: If the risk level is level II or level III, a warning signal of the corresponding level is issued.

[0103] According to the driver information of the human-driven vehicle and the traffic data of the vehicle and the obstacle ahead obtained by the data acquisition module, the data processing module performs feature extraction and data fusion processing. The driver reaction time model based on fuzzy theory is adopted, and the driver age information, driving experience information and driver fatigue level are used as the characteristic values ​​of the input model, and the characteristic values ​​are fuzzified. Through the logical operation of fuzzy rules, a fuzzy quantity of the driver reaction time is obtained, and then the fuzzy quantity is converted into an accurate specific value by defuzzification. The driver reaction time is used as the input of the human-driven vehicle safety distance model, and the safety distance that the vehicle and the vehicle ahead need to maintain is calculated by combining the parameters such as the vehicle speed, the speed of the vehicle ahead, and the maximum deceleration of the vehicle. For unmanned vehicles, according to the obstacle data ahead obtained by multiple sensors, the current speed of the unmanned vehicle and other parameters, the safety distance that the unmanned vehicle needs to maintain between the unmanned vehicle and the obstacle ahead is calculated through the unmanned vehicle safety distance model. The actual measured distance between vehicles is compared with the calculated safety distance, and different collision risk levels are divided according to the size of the distance difference. Determine the collision risk level. If the risk level is level II or level III, the graded warning module will send a warning signal of the corresponding degree to remind the driver to take measures to avoid collision. If the collision risk level reaches level I and the driver of the human-driven vehicle does not respond in time, the decision center sends a control instruction to the anti-collision control system, and the human-driven vehicle control module combines the automatic emergency braking system to brake. For unmanned vehicles, when the collision risk level reaches level I, the unmanned vehicle control module directly performs braking control in combination with the automatic driving system according to the instructions of the decision center, without warning.

[0104] The data acquisition module obtains information such as driver age, driving experience, fatigue level, and traffic data such as vehicle distance and speed through sensors such as on-board cameras and millimeter-wave radars. The data processing module uses convolutional neural networks to extract features from the acquired image data, extract key information such as driver facial features and vehicle contours, and fuse the data of different sensors through data fusion algorithms to obtain accurate traffic status information. The driver reaction time model is based on fuzzy theory, and fuzzifies the characteristic values ​​of driver age, driving experience, fatigue level, etc. The age is divided into three fuzzy sets of "youth", "middle-aged", and "old", the driving experience is divided into three fuzzy sets of "novice", "skilled", and "veteran", and the fatigue level is divided into three fuzzy sets of "awake", "mild fatigue", and "severe fatigue". By establishing fuzzy rules, such as "if the driver is middle-aged, the driving experience is veteran, and the fatigue level is mild fatigue, then the reaction time is faster", the maximum membership method is used for fuzzy reasoning, and the fuzzy quantity of the driver's reaction time is obtained, and then the center of gravity method is used for defuzzification to convert the fuzzy quantity into a specific reaction time value, such as 8 seconds. The safe distance model of human-driven vehicles takes parameters such as reaction time, vehicle speed, front vehicle speed, and maximum deceleration as input, and calculates the safe distance through the formula d=vt+v^2 / (2a), where d is the safe distance, v is the vehicle speed, t is the driver's reaction time, and a is the maximum deceleration. The safe distance model of unmanned vehicles takes parameters such as obstacle distance and vehicle speed as input, and calculates the safe distance through dynamic models and optimization algorithms. For example, the model predictive control method is used to predict the vehicle's motion trajectory in the future and solve the optimal control sequence to make the vehicle decelerate smoothly within the safe distance. The collision risk assessment module compares the actual vehicle distance with the safe distance. If the difference is less than 2 meters, it is a level I risk, 2-5 meters is a level II risk, 5-10 meters is a level III risk, and greater than 10 meters is safe. When the risk level is level II or level III, the graded warning module sends a warning signal through the indicator light on the dashboard or voice prompts to remind the driver to keep a safe distance. When the risk level is level I and the driver fails to take timely measures, the anti-collision control system takes over the vehicle, controls the electronic stability system, brake booster and other actuators, implements automatic emergency braking, and decelerates the vehicle at a deceleration of 8g to avoid a collision. When the risk level is level I, the unmanned vehicle directly executes the braking control command issued by the decision center, controls the steering motor, brake motor and other actuators, and stops the vehicle with the shortest braking distance without manual intervention.

[0105] The driver's facial features are extracted through a convolutional neural network to obtain the driver's fatigue level information.

[0106] The fuzzy theory is used to fuzzy the driver's age, driving experience and fatigue level to determine the fuzzy value of the driver's reaction time.

[0107] The safe distance of the driverless vehicle is calculated through the dynamic model and optimization algorithm, and the optimal control sequence under the vehicle's future motion trajectory is determined.

[0108] Step S108: If the risk level is level I and the driver does not respond in time, emergency braking is performed on the human-driven vehicle, and deceleration or braking control is performed on the unmanned vehicle.

[0109] Obtain the driver information of the human-driven vehicle and the traffic data of the current vehicle and the vehicle in front during driving, and establish different safety distance models according to the human-driven vehicle and the unmanned vehicle respectively; analyze and process the acquired data, divide the collision risk level according to the risk level index, and obtain the collision risk level result after division; determine whether the collision risk level is level I risk, if so, proceed to step 4; if not, proceed to step 7; determine whether the driver of the human-driven vehicle responds in time, if so, proceed to step 7; if not, proceed to step 5; according to the judgment result of step 4, if If the driver of the human-driven vehicle fails to respond in time, the automatic emergency braking system will be used to implement emergency braking on the human-driven vehicle; according to the judgment result of step 3, if the collision risk level is level I risk, the speed or braking of the unmanned vehicle will be controlled according to the risk level, and then step 9 will be entered; according to the judgment result of step 3, if the collision risk level is level II risk or level III risk, the collision warning module will issue a corresponding level II or level III warning; if the collision risk level is not level I, level II or level III risk, return to step 1 to continue to obtain data and analyze and process; and end the collision warning and control process.

[0110] Through on-board sensors, cameras and other equipment, the physiological information of human-driven vehicle drivers (such as heart rate, breathing rate, etc.) and traffic data such as speed, acceleration, and distance during vehicle driving are obtained in real time. According to the different characteristics of human-driven vehicles and unmanned vehicles, safety distance models are established respectively, such as the safety distance of human-driven vehicles d1 = v1 × t1 + v1 2 / (2×a1), the safe distance of the driverless vehicle d2=v2×t2+v2 2 / (2×a2), where v is the vehicle speed, t is the reaction time, and a is the braking acceleration. The acquired data is analyzed and processed by the fuzzy comprehensive evaluation method. Combined with vehicle speed, vehicle distance, driver status and other indicators, the membership function is calculated to obtain the membership values ​​of the collision risk belonging to level Ⅰ (emergency), level Ⅱ (dangerous), and level Ⅲ (general). The maximum value is taken to determine the risk level. When the collision risk is level Ⅰ, if the driver's reaction time exceeds 5 seconds, the automatic emergency braking system intervenes and applies a maximum braking force of 8g to the vehicle. At the same time, the unmanned vehicle reduces the speed or applies the brake according to the risk level. For example, when the risk is level Ⅰ, the speed is reduced by 20% and the braking force is 6g. When the collision risk is level Ⅱ or level Ⅲ, the collision warning module warns the driver by means of beeps, voice prompts, etc. If the risk level is lower than level Ⅲ, continue to obtain data and conduct the next round of analysis. Through the cyclic execution of the above process, the continuous monitoring and control of the collision risk in a mixed traffic environment is realized.

[0111] The driver's physiological information and traffic data are obtained in real time through on-board sensors and cameras to obtain an accurate collision risk level.

[0112] The above is only a preferred embodiment of the present invention. Therefore, any equivalent changes or modifications made according to the structure, characteristics and principles described in the scope of the patent application of the present invention are included in the protection scope of the patent application of the present invention.

Claims

1. A vehicle forward collision warning and control system suitable for mixed human and machine driving traffic flow, characterized in that: Including decision-making center and collision warning system and collision avoidance control system; The collision warning system collects information about the driver of a human-driven vehicle and traffic data about the current vehicle and the vehicle ahead during driving. It establishes different safety distance models for human-driven vehicles and unmanned vehicles respectively, and divides the collision risk level according to the risk level index and issues graded warnings based on the data analysis results. The divided results are sent to the decision-making center. The decision center determines whether to control the vehicle based on the collision risk level and sends the decision result to the collision avoidance control system; The anti-collision control system controls the vehicle. When the human-driven vehicle fails to respond to the warning in time, that is, when the collision risk level becomes high, the automatic emergency braking system will be used to apply emergency braking to the vehicle, and the speed or brake control of the unmanned vehicle will be carried out according to the risk level.

2. The vehicle forward collision warning and control system applicable to mixed human-machine driving traffic flow according to claim 1, characterized in that: The collision warning system includes a data acquisition module, a data analysis module and a graded warning module; The data acquisition module is used to collect the driver information of the human-driven vehicle and the data of the vehicle and the obstacles ahead, and obtain the driver information of the human-driven vehicle and the traffic data of the vehicle and the obstacles ahead; The data analysis module processes the traffic data of the vehicle and the obstacles ahead, and classifies the risks accordingly; The graded warning module provides different degrees of warning based on different risk levels.

3. The vehicle forward collision warning and control system applicable to mixed human-machine driving traffic flow according to claim 2, characterized in that: The obstacle data ahead are data measured by solid-state laser radar and millimeter-wave radar respectively; the driver information of human-driven vehicles includes driver age information, driving experience information, and driver fatigue level information.

4. The vehicle forward collision warning and control system applicable to mixed human-machine driving traffic flow according to claim 3 is characterized in that: The data analysis module integrates the obstacle data ahead and fuzzifies the driver information of the human-driven vehicle, including the driver's age information, driving experience information, and driver fatigue information, and estimates the driver's reaction time through fuzzy logic reasoning to obtain the collision risk level. The process includes: Extract features from the data to obtain features related to the safety distance model; through feature input, obtain a safety distance model for human-driven vehicles and a safety distance model for unmanned vehicles based on driver characteristics; The fused data is compared with its corresponding safety distance, and the collision risk is graded using the collision risk index.

5. The vehicle forward collision warning and control system applicable to mixed human-machine driving traffic flow according to claim 3 is characterized in that: The process of the graded warning module for graded warnings of different risk levels includes: According to the classification of collision risk levels, warnings of corresponding degrees are given.

6. The vehicle forward collision warning and control system applicable to mixed human-machine driving traffic flow according to claim 4, characterized in that: The process of the data analysis module classifying the risk level according to the collision risk index based on the fused data includes: Construct a safe distance model for human-driven vehicles and an unmanned vehicle based on driver characteristics, input the fused data into the model, obtain the distance output, and determine the collision risk level according to the risk level index; The risk levels include Level I risk, Level II risk, and Level III risk with gradually increasing safety.

7. The automobile forward collision warning and control system applicable to mixed human-machine driving traffic flow according to claim 6 is characterized in that: If the risk level is level II or level III, the collision warning module will issue a corresponding level II or level III warning; If the risk level is Level I, the human-driven vehicle control module will perform emergency braking in conjunction with the automatic emergency braking system when it receives the control command from the decision center, and the unmanned vehicle control module will perform deceleration or braking control according to the command issued by the decision center.

8. The vehicle forward collision warning and control system applicable to mixed human-machine driving traffic flow according to claim 7, characterized in that: The anti-collision control system includes a human-driven vehicle control module and an unmanned vehicle control module; when the human-driven vehicle fails to respond to the warning in time, that is, the collision risk level becomes high risk, the human-driven vehicle control module combines with the automatic emergency braking system to perform emergency braking on the vehicle, and the unmanned vehicle control module controls the vehicle speed or braking according to the collision risk level of the unmanned vehicle.

9. The vehicle forward collision warning and control system applicable to mixed human-machine driving traffic flow according to claim 4, characterized in that: The classification method after data analysis module acquisition includes: obtaining driver information of human-driven vehicles and traffic data of the current vehicle and the vehicle ahead, the driver information includes age, driving experience and fatigue level, and the traffic data includes vehicle speed and relative distance; Using fuzzy logic reasoning method, the driver's reaction time is estimated according to the driver's age, driving experience and fatigue level; According to the driver's reaction time and vehicle driving status, a safe distance model for human-driven vehicles is established; Use multi-sensor fusion method to obtain obstacle data ahead and establish a safe distance model for unmanned vehicles; Compare the actual vehicle distance with the calculation result of the safety distance model to obtain a collision risk score; The risk levels are divided according to the collision risk score, including level I, level II, and level III risks; If the risk level is II or III, a warning signal of the corresponding level will be issued; If the risk level is Level I and the driver does not respond in time, emergency braking will be applied to human-driven vehicles and deceleration or braking control will be implemented for unmanned vehicles.

Images