A lane departure warning algorithm evaluation method considering lateral and longitudinal safety distances

Abstract

The application relates to a lane departure warning algorithm evaluation method considering lateral and longitudinal safety distances, which is used for judging the safety of an existing lane departure warning algorithm and improving the false alarm rate of the system, and comprises the following steps: extracting a lane departure event from a natural driving database; obtaining lateral and longitudinal safety indexes by respectively calibrating the longitudinal safety distance and the lateral safety distance of an RSS model which comprehensively considers safety and efficiency; and comparing and evaluating the lateral and longitudinal safety distances involved in the lane departure warning algorithm based on the calibrated RSS model. Compared with the prior art, the lateral and longitudinal safety distance indexes constructed by the application comprehensively consider safety and efficiency, can realize quantitative algorithm safety evaluation, are helpful for finding the rationality of the lane departure warning algorithm, can find key safety parameters involved in the algorithm, and improve the precision of the algorithm.

Description

Technical Field

[0001] This invention relates to the field of vehicle active safety function algorithm evaluation technology, and in particular to a lane departure warning algorithm evaluation method that considers lateral and longitudinal safety distances. Background Technology

[0002] Statistics from the National Highway Traffic Safety Administration (NHTSA) show that lane departure accidents account for more than one-third (37.4%) of fatal accidents. To reduce the risk of traffic accidents, lane departure warning systems (LDWs) were developed and widely installed in vehicles.

[0003] However, current LDW (Driver Delay Warning) systems on the market suffer from numerous unnecessary warnings due to their limited application scenarios, fixed warning thresholds, and algorithms that do not consider driver needs. These false alarms result in low driver acceptance of the system. Therefore, there is an urgent need to propose an LDW algorithm evaluation method that aligns with driver behavior characteristics to improve the safety of LDW algorithms and provide a theoretical basis for LDW algorithm standards.

[0004] Current methods for evaluating and testing LDW (Low-Density Warning) systems primarily rely on field tests, such as those conducted by Euro NCAP and NHT SANCAP. These methods typically use fixed thresholds and set parameters for different scenarios (high-speed, low-speed, cornering, etc.) to test the LDW system. Such tests cannot clearly identify the causes of false alarms and false negatives in the system's algorithm, nor can they determine the key parameters affecting the algorithm. Therefore, there is an urgent need to develop a quantitative testing metric to quantitatively assess the system's safety. This would help identify the key parameters affecting the algorithm and provide a basis for algorithm optimization. Summary of the Invention

[0005] The purpose of this invention is to provide an evaluation method for lane departure warning algorithms that takes into account both lateral and longitudinal safety distances.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] An evaluation method for lane departure warning algorithms that considers lateral and longitudinal safety distances includes the following steps:

[0008] Step 1) Extract lane departure events from the natural driving database to obtain warning video data and vehicle motion data;

[0009] Step 2) Divide the lane departure events extracted in Step 1) into single-vehicle lane departure scenarios and multi-vehicle lane departure scenarios;

[0010] Step 3) Use the k-shape-clustering clustering method to cluster the single-vehicle lane departure scenarios and multi-vehicle lane departure scenarios divided in Step 2) respectively, to obtain single-vehicle lane departure high-risk scenarios and single-vehicle lane departure low-risk scenarios, multi-vehicle lane departure high-risk scenarios and multi-vehicle lane departure low-risk scenarios.

[0011] Step 4) Use the lane departure natural driving data of the four clustered scenarios to calibrate the lateral and longitudinal safe distances based on the Responsibility Sensitive Safe (RSS) model.

[0012] Step 5) Evaluate the safety of existing lane departure warning algorithms based on the calibrated lateral and longitudinal safety distances.

[0013] The extraction of lane departure events from the natural driving database specifically involves: setting a threshold for extracting lane departure events, and extracting lane departure events from the natural driving data based on the threshold, wherein the threshold includes a distance threshold, a time threshold, and a speed threshold.

[0014] The vehicle motion data for the lane departure event in step 1) includes: vehicle speed, time-to-lane-crossing (TLC), vehicle speed and relative distance of vehicles in adjacent lanes, vehicle speed and relative distance of the vehicle in front, and vehicle lateral offset.

[0015] In step 2), the variables involved in classifying single-vehicle lane departure scenarios include the vehicle's speed, lateral deviation distance, lateral crossing time, and road type; the variables involved in classifying multi-vehicle lane departure scenarios include the vehicle's speed, lateral deviation distance, lateral crossing time, relative speeds of vehicles in adjacent lanes, relative distances of vehicles in adjacent lanes, speed of the vehicle in front, relative distances of the vehicle in front, and road type; wherein, the road type includes highways, urban expressways, urban roads, and rural roads.

[0016] In step 3),

[0017] Among them, the characteristics of high-risk scenarios for single-vehicle lane departure are: increased vehicle speed, lateral distance first decreasing then increasing and then decreasing again, and TLC first decreasing and then increasing;

[0018] The characteristics of low-risk lane departure scenarios for single vehicles are: reduced vehicle speed, reduced lateral distance, and reduced TLC;

[0019] The characteristics of high-risk multi-vehicle lane departure scenarios are: increasing vehicle speed, lateral distance first decreasing then increasing and then decreasing again, and TLC first decreasing and then increasing;

[0020] The characteristics of low-risk multi-vehicle lane departure scenarios are: reduced vehicle speed, decreased lateral distance, and TLC (Traffic Lane Departure) initially decreasing and then increasing.

[0021] In step 4), the lateral safety distance based on the RSS model is expressed as:

[0022]

[0023] in, Let v1 be the lateral safety distance, v2 be the lateral speed of the vehicle in the adjacent lane, μ be the final lateral safety distance, and ρ be the lateral reaction time. For the minimum lateral comfort deceleration of the vehicle; μ, ρ, These are the parameters that need to be calibrated.

[0024] In step 4), the method for parameter calibration in the lateral safety distance based on the RSS model is the NSGA-II algorithm, and the optimization objective is:

[0025] Objective 1:

[0026] Objective 2:

[0027]

[0028] Where TLC is the lateral crossing time, TLC* is 2.7s, and TLCi is the lateral crossing time at time i.

[0029] In step 4), the longitudinal safety distance based on the RSS model is expressed as:

[0030]

[0031] in, For longitudinal safety distance, v r v is the longitudinal velocity of the vehicle. f Let ρ be the longitudinal velocity of the vehicle in front, ρ be the lateral reaction time, and a be the longitudinal velocity of the vehicle in front. max，accel Let a be the maximum acceleration of the vehicle in front. min，brake、 a max，brake These represent the minimum comfort and maximum deceleration of the vehicle in front, respectively.

[0032] In step 4), the method for parameter calibration in the longitudinal safety distance based on the RSS model is the NSGA-II algorithm, and the optimization objective is:

[0033] Objective 1:

[0034]

[0035] Objective 2:

[0036]

[0037] Where TIT is the longitudinal safety time integral, TTC* is 4.5s, and TIT i Let be the safe time at time i.

[0038] Step 5) specifically involves: evaluating the actual lateral and longitudinal distances at the warning time of the existing lane departure warning algorithm based on the calibrated lateral and longitudinal safety distances, and using a confusion matrix to calculate the accuracy and false alarm rate of the lane departure warning algorithm to assess its safety.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] (1) Quantitative evaluation algorithm: The horizontal and vertical safety distances in this invention can quantitatively evaluate the safety, false alarm rate and accuracy of the existing LDW algorithm, making up for the shortcomings of the qualitative evaluation of previous algorithms.

[0041] (2) Comprehensive evaluation indicators: The horizontal and vertical safety distance calculation model in this invention involves driver behavior characteristics such as reaction time, braking force, and vehicle speed, which is closer to the driver behavior characteristics' requirements for the LDW system. Attached Figure Description

[0042] Figure 1 This is a flowchart of the method of the present invention;

[0043] Figure 2 This is a schematic diagram illustrating the classification of lane departure scenarios in an embodiment of the present invention. Detailed Implementation

[0044] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0045] This embodiment provides a lane departure warning algorithm evaluation method that considers both lateral and longitudinal safety distances, such as... Figure 1 As shown, it includes the following steps:

[0046] Step 1) Extract lane departure events from the natural driving database to obtain warning video data and vehicle motion data.

[0047] Specifically, a threshold is set for extracting lane departure events. Lane departure events are extracted from natural driving data based on this threshold. Events are considered lane departure events when they meet the following conditions: distance threshold (distance between vehicle centerline and lane line ∈ [0.9, 1.025] m or distance between lane centerline and lane line ∈ [-1.75, 0.9) m], time threshold (lane departure duration > 3 seconds), and speed threshold (vehicle speed > 5 m / s). In this embodiment, the extracted lane departure video and vehicle motion data are 10 seconds of continuous video and continuous data. The vehicle motion data includes:

[0048] ① Vehicle speed;

[0049] ②Time-to-Lane-Crossing (TLC);

[0050] ③Speed ​​and relative distance of vehicles in adjacent lanes;

[0051] ④Speed ​​of the vehicle in front and relative distance;

[0052] ⑤ Vehicle lateral offset.

[0053] This embodiment filters lane departure events from natural driving data, totaling 516 incidents. To ensure data accuracy, this embodiment also incorporates manual video observation, data cleaning, and outlier removal.

[0054] Step 2) Divide the lane departure events extracted in Step 1) into single-vehicle lane departure scenarios and multi-vehicle lane departure scenarios.

[0055] Specifically, the variables involved in classifying single-vehicle lane departure scenarios include vehicle speed, lateral offset distance, TLC (Traffic Carrier Flow), and road type (highway, urban expressway, urban road, rural road).

[0056] The variables involved in classifying multi-vehicle lane departure scenarios include the vehicle's speed, lateral offset distance, TLC, relative speeds of vehicles in adjacent lanes, relative distances of vehicles in adjacent lanes, speed of the vehicle in front, relative distances of the vehicle in front, and road type (highway, urban expressway, urban road, rural road).

[0057] Step 3) Use the k-shape-clustering method to cluster the single-vehicle lane departure scenarios and multi-vehicle lane departure scenarios divided in Step 2), respectively, to obtain high-risk and low-risk single-vehicle lane departure scenarios, and high-risk and low-risk multi-vehicle lane departure scenarios, such as... Figure 2 As shown.

[0058] Among them, the characteristics of high-risk scenarios for single-vehicle lane departure are: increased vehicle speed, lateral distance first decreasing then increasing and then decreasing again, and TLC first decreasing and then increasing;

[0059] The characteristics of low-risk lane departure scenarios for single vehicles are: reduced vehicle speed, reduced lateral distance, and reduced TLC;

[0060] The characteristics of high-risk multi-vehicle lane departure scenarios are: increasing vehicle speed, lateral distance first decreasing then increasing and then decreasing again, and TLC first decreasing and then increasing;

[0061] The characteristics of low-risk multi-vehicle lane departure scenarios are: reduced vehicle speed, decreased lateral distance, and TLC (Traffic Lane Departure) initially decreasing and then increasing.

[0062] Step 4) Use the lane departure natural driving data of the four clustered scenarios to calibrate the lateral and longitudinal safe distances based on the Responsibility Sensitive Safe (RSS) model.

[0063] The lateral safety distance based on the RSS model is expressed as:

[0064]

[0065] in, Let v1 be the lateral safety distance, v2 be the lateral speed of the vehicle in the adjacent lane, μ be the final lateral safety distance, and ρ be the lateral reaction time. For the minimum lateral comfort deceleration of the vehicle; μ, ρ, These are the parameters that need to be calibrated.

[0066] The method for parameter calibration in the lateral safety distance based on the RSS model is the NSGA-II (Non-dominated Sorting Genetic Algorithm-II) algorithm, and the optimization objective is:

[0067] Objective 1:

[0068] Objective 2:

[0069]

[0070] TLC is the lateral crossing time, TLC* is 2.7s, and TLCi is the lateral crossing time at time i.

[0071] Objective 1 is to ensure the safety of the algorithm, and objective 2 is to ensure the efficiency of the algorithm.

[0072] The longitudinal safety distance based on the RSS model is expressed as:

[0073]

[0074] in, For longitudinal safety distance, v r v is the longitudinal velocity of the vehicle. f Let ρ be the longitudinal velocity of the vehicle in front, ρ be the lateral reaction time, and a be the longitudinal velocity of the vehicle in front. max，accel Let a be the maximum acceleration of the vehicle in front. min，brake a max，brake These represent the minimum comfort and maximum deceleration of the vehicle in front, respectively.

[0075] The parameter calibration method for longitudinal safety distance based on the RSS model is the NSGA-II algorithm, and the optimization objective is:

[0076] Objective 1:

[0077]

[0078] Objective 2:

[0079]

[0080] TIT is the longitudinal safety time integral, TTC* is 4.5s, TIT i Let be the safe time at time i.

[0081] Objective 1 is to ensure the safety of the algorithm, and objective 2 is to ensure the efficiency of the algorithm.

[0082] In this embodiment, the calibrated lateral safety distance is:

[0083] Single-vehicle scenario A (high-risk scenario of lane departure for single-vehicle):

[0084]

[0085] Single-vehicle scenario B (low-risk lane departure scenario):

[0086]

[0087] Multi-vehicle scenario A (high-risk lane departure scenario with multiple vehicles):

[0088]

[0089] Multi-vehicle scenario B (low-risk lane departure scenario for multi-vehicle vehicles):

[0090]

[0091] The calibrated longitudinal safety distance is:

[0092] Multi-vehicle scenario A (high-risk lane departure scenario with multiple vehicles):

[0093]

[0094] Multi-vehicle scenario B (low-risk lane departure scenario for multi-vehicle vehicles):

[0095]

[0096] Step 5) Evaluate the safety of existing lane departure warning algorithms based on the calibrated lateral and longitudinal safety distances.

[0097] Step 5) specifically involves: evaluating the actual lateral and longitudinal distances at the warning time of the existing lane departure warning algorithm based on the calibrated lateral and longitudinal safety distances, and using a confusion matrix to calculate the accuracy, false alarm rate, and other indicators of the lane departure warning algorithm to assess its safety.

[0098] This embodiment uses the calibrated horizontal and vertical safety distance models to evaluate the LDW algorithms of NHTSA and Mobileye, and the results are shown in Table 1.

[0099] Table 1 Evaluation results of the LDW algorithm

[0100]

[0101] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for evaluating lane departure warning algorithms that considers lateral and longitudinal safety distances, characterized in that, Includes the following steps: Step 1) Extract lane departure events from the natural driving database to obtain warning video data and vehicle motion data; Step 2) Divide the lane departure events extracted in Step 1) into single-vehicle lane departure scenarios and multi-vehicle lane departure scenarios; Step 3) Use the k-shape-clustering clustering method to cluster the single-vehicle lane departure scenarios and multi-vehicle lane departure scenarios divided in Step 2) respectively, to obtain single-vehicle lane departure high-risk scenarios and single-vehicle lane departure low-risk scenarios, multi-vehicle lane departure high-risk scenarios and multi-vehicle lane departure low-risk scenarios. Step 4) Use the lane departure natural driving data of the four clustered scenarios to calibrate the lateral and longitudinal safety distances based on the RSS model; Step 5) Evaluate the safety of existing lane departure warning algorithms based on the calibrated lateral and longitudinal safety distances; In step 4), the lateral safety distance based on the RSS model is expressed as: in, For lateral safety distance, The lateral speed of the vehicle, The lateral speed of the adjacent lane. For the final lateral safety distance, For the lateral reaction time, Minimum lateral comfort deceleration for the vehicle; , , These are the parameters that need to be calibrated; In step 4), the method for parameter calibration in the lateral safety distance based on the RSS model is the NSGA-II algorithm, and the optimization objective is: Where TLC is the lateral travel time, TLC As a default value, TLC i For the first i A horizontal journey through time.

2. The lane departure warning algorithm evaluation method considering lateral and longitudinal safety distances according to claim 1, characterized in that, The extraction of lane departure events from the natural driving database specifically involves: setting a threshold for extracting lane departure events, and extracting lane departure events from the natural driving data based on the threshold, wherein the threshold includes a distance threshold, a time threshold, and a speed threshold.

3. The lane departure warning algorithm evaluation method considering lateral and longitudinal safety distances according to claim 1, characterized in that, The vehicle motion data for the lane departure event in step 1) includes: vehicle speed, lateral crossing time, vehicle speed and relative distance of vehicles in adjacent lanes, vehicle speed and relative distance of the vehicle in front, and vehicle lateral offset.

4. The lane departure warning algorithm evaluation method considering lateral and longitudinal safety distances according to claim 1, characterized in that, In step 2), the variables involved in classifying single-vehicle lane departure scenarios include the vehicle's speed, lateral deviation distance, lateral crossing time, and road type; the variables involved in classifying multi-vehicle lane departure scenarios include the vehicle's speed, lateral deviation distance, lateral crossing time, relative speeds of vehicles in adjacent lanes, relative distances of vehicles in adjacent lanes, speed of the vehicle in front, relative distances of the vehicle in front, and road type; wherein, the road type includes highways, urban expressways, urban roads, and rural roads.

5. The lane departure warning algorithm evaluation method considering lateral and longitudinal safety distances according to claim 1, characterized in that, In step 3), Among them, the characteristics of high-risk scenarios for single-vehicle lane departure are: increased vehicle speed, lateral distance first decreasing then increasing and then decreasing again, and lateral crossing time first decreasing and then increasing; The characteristics of low-risk lane departure scenarios for single vehicles are: reduced vehicle speed, reduced lateral distance, and reduced lateral crossing time. The characteristics of high-risk multi-vehicle lane departure scenarios are: increasing vehicle speed, lateral distance first decreasing then increasing and then decreasing again, and TLC first decreasing and then increasing; The characteristics of low-risk multi-vehicle lane departure scenarios are: reduced vehicle speed, reduced lateral distance, and lateral crossing time that first decreases and then increases.

6. The lane departure warning algorithm evaluation method considering lateral and longitudinal safety distances according to claim 1, characterized in that, In step 4), the longitudinal safety distance based on the RSS model is expressed as: in, For longitudinal safety distance, For the longitudinal speed of the vehicle, The longitudinal speed of the vehicle in front. For the lateral reaction time, The maximum acceleration of the vehicle in front. , These represent the minimum comfort and maximum deceleration of the vehicle in front, respectively.

7. The lane departure warning algorithm evaluation method considering lateral and longitudinal safety distances according to claim 6, characterized in that, In step 4), the method for parameter calibration in the longitudinal safety distance based on the RSS model is the NSGA-II algorithm, and the optimization objective is: Objective 1: min Objective 2: Where TIT is the longitudinal safety time integral, and TTC is the safety time integral. TIT is the default value. i For the first i Safety time and time at all times.

8. The lane departure warning algorithm evaluation method considering lateral and longitudinal safety distances according to claim 1, characterized in that, Step 5) specifically involves: evaluating the actual lateral and longitudinal distances at the warning time of the existing lane departure warning algorithm based on the calibrated lateral and longitudinal safety distances, and using a confusion matrix to calculate the accuracy and false alarm rate of the lane departure warning algorithm to assess its safety.

Images