A method for assisting in predicting pedestrian trajectories for autonomous driving

Abstract

This invention discloses a method for predicting pedestrian trajectories in assisted driving, comprising the following steps: Step 1: Acquiring and updating the historical trajectory of a pedestrian target through an onboard imaging device; Step 2: Performing coordinate transformation and data filtering on the historical trajectory; Step 3: Performing curve fitting on the filtered data to obtain the curved trajectory of the pedestrian target; Step 4: Performing Kalman filtering iterative calculation in real time to obtain the Kalman filtered velocity of the pedestrian target; Step 5: Using the obtained Kalman filtered velocity as the moving speed of the pedestrian target, and using the fitted curve as the direction of travel of the pedestrian target, deriving the trajectory of the pedestrian target, and predicting the direction and speed of travel of the pedestrian target in real time. The method of this invention, based on a combination of historical trajectory fitting and Kalman filtering, achieves real-time prediction of the pedestrian target's trajectory.

Description

Technical Field

[0001] This invention relates to the field of driver assistance technology, and more specifically to a method for predicting pedestrian trajectories in driver assistance systems. Background Technology

[0002] In traffic scenarios where assisted driving is used, pedestrians generally do not interact much with the vehicle. Although pedestrians' movements are flexible, unpredictable, and difficult to anticipate, it is sufficient to focus on two types of pedestrian movements that may have a significant impact on the vehicle: pedestrians crossing the road and pedestrians standing still on the side of the road or walking slowly along the road. Predicting these two behaviors requires observing the historical state of pedestrians, analyzing their movement trends through fitting, statistics, and other methods, deriving their intentions, and further predicting their trajectories based on their intentions.

[0003] Patent CN202110433242X discloses a method, device, equipment, and storage medium for predicting pedestrian trajectories. This patent utilizes an LSTM network for training. LSTM is a deep learning network model that can learn historical information over a certain period. However, this type of technology suffers from drawbacks such as reliance on data training, lack of interpretability, long software mass production cycles, and a lack of effective means to adjust when prediction results contradict expectations. Furthermore, the long mass production cycle and unpredictable results make it difficult to ensure learning effectiveness.

[0004] Patent CN110606019A discloses a method, device, vehicle, and storage medium for pedestrian yielding in autonomous driving. This patent relies entirely on manually set rules and subjective judgment for prediction. The method is relatively simple; for example, it mainly considers whether a pedestrian is moving towards the lane. If they are moving towards the lane, it is highly likely they will cross the lane. This is combined with the pedestrian's distance from the center of the lane to make a confident judgment about their intention to cross. Based on this intention judgment, a simple uniform velocity model is used to calculate the trajectory of the pedestrian from their current position, taking into account their current speed and direction. However, this technology is relatively crude, relying too heavily on the lateral velocity of the pedestrian target obtained from sensors, which suffers from significant detection instability. Summary of the Invention

[0005] In view of the above-mentioned shortcomings of the existing technology, the purpose of the present invention is to provide a method for predicting pedestrian trajectories in assisted driving, so as to solve the problems of existing technologies that rely on data training for pedestrian trajectory prediction, lack effective means to adjust when the prediction results contradict expectations, and rely too much on the lateral velocity of pedestrian targets obtained by sensors.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] A method for predicting pedestrian trajectories in assisted driving includes the following steps:

[0008] Step 1: Acquire and update the historical trajectory of pedestrian targets using vehicle-mounted imaging equipment;

[0009] Step 2: Perform coordinate transformation and data filtering on the historical trajectories obtained in Step 1;

[0010] Step 3: Perform curve fitting on the data filtered in Step 2 to obtain the curve trajectory of the pedestrian target;

[0011] Step 4: Use Kalman filtering iterative calculation to perform Kalman filtering calculation in real time to obtain the Kalman filtered velocity of the pedestrian target;

[0012] Step 5: Using the Kalman filter velocity obtained in Step 4 as the moving speed of the pedestrian target and the fitted curve obtained in Step 3 as the direction of travel of the pedestrian target, the trajectory of the pedestrian target is derived, and the direction and speed of travel of the pedestrian target are predicted in real time.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] 1. The method described in this invention is based on a combination of historical trajectory fitting and Kalman filtering. The Kalman filter is used to derive the moving speed of the pedestrian target in the next frame. At the same time, the curve fitting of the historical trajectory is used to obtain the trajectory direction of the pedestrian target. Combined with the moving speed, the real-time prediction of the pedestrian target trajectory is realized.

[0015] 2. The method described in this invention can predict the direction of travel and speed of pedestrian targets in real time without relying on a large amount of data for training. When the prediction result contradicts the expectation, it can analyze the erroneous data in a timely and accurate manner, thereby optimizing the method in a timely manner. At the same time, it reduces the dependence on sensors and improves the stability of predicting the lateral speed of pedestrian targets. Moreover, even if a pedestrian suddenly turns while walking, the method described in this invention can adjust the direction of travel in time when the pedestrian turns and predict the pedestrian trajectory in real time based on the new direction of travel, so that vehicles can avoid the pedestrian in time on the driving map. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for predicting pedestrian trajectories in assisted driving according to the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] This invention provides a method for predicting pedestrian trajectories in assisted driving, such as... Figure 1 As shown, it includes the following steps:

[0019] Step 1: Acquire and update the historical trajectory of pedestrian targets using the vehicle-mounted imaging device. In Step 1, the historical trajectory of the pedestrian targets consists of the historical position coordinate information of multiple pedestrian targets, where the coordinate information is the relative lateral and longitudinal position based on the vehicle's coordinate system. In practice, the historical trajectory of the pedestrian targets is continuously acquired and updated. A fixed-length array stores the position coordinates of the pedestrian targets and updates it with each time step, ensuring that the values ​​in the array represent the historical position coordinate information of the pedestrian targets.

[0020] Step 2: Perform coordinate transformation and data filtering on the historical trajectory obtained in Step 1. In Step 2, the coordinate transformation is performed by sequentially offsetting the vehicle's motion drift frame by frame. The data filtering involves checking the latest consecutive non-zero segment within the historical position coordinate information of pedestrian targets for prediction purposes. In actual use, since the vehicle is constantly moving, the historical position coordinate information of pedestrian targets is based on the relative lateral and longitudinal distances of the vehicle's coordinate system, which cannot accurately represent the vehicle's historical trajectory relative to the ground. Therefore, coordinate transformation is necessary. Specifically, since the vehicle is in motion, and the lateral and longitudinal distances are calculated based on the vehicle's relative position, the vehicle's motion will cause coordinate system drift. Coordinate transformation is essentially performing sequential frame-step offsetting of the vehicle's motion drift. Data filtering mainly involves checking the latest consecutive non-zero segment within the array for prediction purposes.

[0021] Step 3: Perform curve fitting on the data filtered in Step 2 to obtain the curve trajectory of the pedestrian target. In Step 3, select the latest continuous non-zero historical coordinate information of the pedestrian target, perform multiple curve fittings to obtain the fitting parameters, and calculate the evaluation value R of the fitting effect. 2 To eliminate the impact of abnormally missing points, based on the data filtering in step 2, the latest continuous and non-zero historical trajectory of the pedestrian target is selected, and cubic curve fitting is performed to obtain the fitting parameters. Simultaneously, the evaluation value R of the fitting effect is calculated. 2 .

[0022] Step 4: Use Kalman filtering iterative calculation to perform Kalman filtering calculation in real time to obtain the Kalman filtered velocity of the pedestrian target.

[0023] Step 5: Using the Kalman-filtered velocity obtained in Step 4 as the pedestrian target's moving speed, and the fitted curve obtained in Step 3 as the pedestrian target's direction of travel, the pedestrian target's trajectory is derived, and the pedestrian target's direction of travel and velocity are predicted in real time. In Step 5, analogical calculus is used to derive the pedestrian target's trajectory. The specific steps are as follows:

[0024] (1) Starting from the current point, take multiple points of equal interval or equal length on the line extending from the curve trajectory, set the prediction time, and multiply it by the Kalman filter speed to obtain the total distance to be traveled.

[0025] (2) Calculate the distance between the first point and the second point, and use this distance as the equivalent trajectory length; if the distance is less than the total distance, continue to calculate the distance between the second point and the third point, and add it to the distance between the first point and the second point; if the result after adding is still less than the total distance, continue to add the next segment.

[0026] (3) After superimposing the distance between the nth point and the (n+1)th point, if the superimposed result is greater than the total distance, then the distance between the nth point and the (n-1)th point and the distance between the nth point and the (n+1)th point are judged, and the point with the smallest distance is taken as the endpoint. The curve segment between the starting point and the endpoint is output as the predicted trajectory.

[0027] Specifically, using an analogy to calculus, starting from the current point, a series of points with equal intervals or lengths are taken along the extended curve trajectory. Then, a prediction time is set, such as 8 seconds, and the total distance to be traveled is obtained by multiplying by the Kalman filter speed. Next, the distance between the first and second points is calculated, and the point interval is reduced so that the straight-line distance can be used as the equivalent trajectory length. If the first straight-line length does not reach the travel distance, the distance between the second and third points is calculated and added to the first distance. If the travel distance is still not reached, the next segment is added until a certain segment is added. After the travel distance is exceeded, the distance to the previous segment or the next segment is calculated, and the closest point is taken as the end point of the travel distance. The curve segment between the starting point and the end point is the predicted trajectory.

[0028] As described above, the reminder system of the present invention is not limited to the configuration described herein, and other systems that can implement the embodiments of the present invention can fall within the scope of protection of the present invention.

[0029] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit the technical solutions. Those skilled in the art should understand that any modifications or equivalent substitutions to the technical solutions of the present invention without departing from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting pedestrian trajectories in assisted driving, characterized in that, Includes the following steps: Step 1: Acquire and update the historical trajectory of pedestrian targets using vehicle-mounted imaging equipment; Step 2: Perform coordinate transformation and data filtering on the historical trajectories obtained in Step 1; Step 3: Perform curve fitting on the data filtered in Step 2 to obtain the curve trajectory of the pedestrian target; Step 4: Use Kalman filtering iterative calculation to perform Kalman filtering calculation in real time to obtain the Kalman filtered velocity of the pedestrian target; Step 5: Use the Kalman filter velocity obtained in Step 4 as the moving speed of the pedestrian target in the next frame, and use the curve trajectory obtained by fitting the historical trajectory curve in Step 3 as the trajectory direction of the pedestrian target. Combine the moving speed to realize real-time prediction of the pedestrian target trajectory.

2. The method for predicting pedestrian trajectories in assisted driving according to claim 1, characterized in that, In step 1, the historical trajectory of the pedestrian target is the historical position coordinate information of multiple pedestrian targets, and the coordinate information is the relative horizontal and vertical position based on the vehicle coordinate system.

3. The method for predicting pedestrian trajectories in assisted driving according to claim 1, characterized in that, In step 2, the coordinates are converted into sequential frame steps to offset the motion drift of the vehicle.

4. The method for predicting pedestrian trajectories in assisted driving according to claim 1, characterized in that, In step 2, the data filtering involves checking the latest continuous non-zero segment within the historical location coordinate information of the pedestrian target for prediction purposes.

5. The method for predicting pedestrian trajectories in assisted driving according to claim 4, characterized in that, In step 3, the latest segment of continuous non-zero historical coordinate information of the pedestrian target is selected, and multiple curve fittings are performed to obtain the fitting parameters. At the same time, the evaluation value R of the fitting effect is calculated. 2 .

6. The method for predicting pedestrian trajectories in assisted driving according to claim 1, characterized in that, In step 5, analogical calculus is used to derive the trajectory of the pedestrian target. The specific steps for deriving the trajectory of the pedestrian target are as follows: (1) Starting from the current point, take multiple points of equal interval or equal length on the line extending from the curve trajectory, set the prediction time, and multiply it by the Kalman filter speed to obtain the total distance to be traveled. (2) Calculate the distance between the first point and the second point, and use this distance as the equivalent trajectory length; if the distance is less than the total distance, continue to calculate the distance between the second point and the third point, and add it to the distance between the first point and the second point; if the result after adding is still less than the total distance, continue to add the next segment. (3) After superimposing the distance between the nth point and the (n+1)th point, if the superimposed result is greater than the total journey, then the distance between the nth point and the (n-1)th point and the distance between the nth point and the (n+1)th point are judged, and the point closest to the starting point is taken as the end point of the journey. The curve segment between the starting point and the end point is output as the predicted trajectory.

Images