Pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning

Abstract

The invention discloses a pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning, and the method comprises the steps: pedestrian target detection, pedestrian trajectory tracking, pedestrian trajectory prediction, vehicle collision avoidance control based on pedestrian trajectory prediction data, and vehicle collision avoidance decision output. According to the target detection and trajectory prediction method in combination with deep learning, pedestrians can be accurately identified and future motion trajectories of the pedestrians can be predicted, so that an intelligent vehicle can adjust a driving strategy in advance, traffic accidents are effectively avoided, and the safety and stability of an automatic driving system are improved.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent driving, and particularly relates to a pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning. Background Art

[0002] Existing intelligent driving systems still face many challenges in the problem of pedestrian collision avoidance in complex urban environments:

[0003] First of all, pedestrian trajectories have great randomness, and traditional methods based on linear prediction or Bayesian inference are difficult to accurately predict pedestrian trajectories; in addition, the existing object detection methods are limited in accuracy in occluded, low-light or dense pedestrian environments, affecting the vehicle's recognition of pedestrian intentions.

[0004] Existing traditional collision avoidance strategies are mostly based on fixed rules or simple PID control, which are difficult to adapt to different pedestrian behavior patterns and are prone to misjudgment or collision avoidance delay;

[0005] Therefore, there is an urgent need for a pedestrian trajectory prediction and intelligent collision avoidance control method combined with deep learning to improve the safety and robustness of the autonomous driving system. Summary of the Invention

[0006] Aiming at the above problems, the purpose of the present invention is to provide a pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning.

[0007] The specific technical solution to achieve the purpose of the present invention is as follows:

[0008] A pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning, comprising the following steps:

[0009] Step 1, perform pedestrian target detection;

[0010] Step 2, perform pedestrian trajectory tracking based on the pedestrian target detection result;

[0011] Step 3, perform pedestrian trajectory prediction based on the pedestrian trajectory tracking result;

[0012] Step 4, perform vehicle collision avoidance control based on the pedestrian trajectory prediction data and output a vehicle collision avoidance decision.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0014] (1) The solution of the present invention can improve the pedestrian detection accuracy: using YOLOv8+CBAM for target detection, compared with the traditional YOLO model, the detection accuracy for small target pedestrians and complex backgrounds is improved by about 5% - 7%;

[0015] (2) The solution of the present invention can enhance the stability of trajectory tracking: The DeepSORT+Re-ID mechanism ensures that pedestrians can still be correctly matched after a short occlusion, avoiding ID loss and mis-tracking;

[0016] (3) The solution of the present invention can optimize the accuracy of trajectory prediction: Social-LSTM models the interaction behavior of pedestrians through a social pooling mechanism, reducing the trajectory prediction error by 15% - 20% compared with the traditional LSTM method;

[0017] (4) The solution of the present invention can improve the robustness of the collision avoidance strategy: The combination of MPC lateral control + PID longitudinal control and Social-LSTM trajectory prediction makes the vehicle collision avoidance strategy more intelligent and improves driving safety.

[0018] The following further describes the present invention in conjunction with specific embodiments. Brief Description of the Drawings

[0019] Figure 1 It is a schematic flow chart of the pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning of the present invention.

[0020] Figure 2 It is a schematic diagram of the YOLOV8 network structure for object detection of the present invention.

[0021] Figure 3 It is a schematic diagram of the CBAM attention mechanism network structure for object detection of the present invention.

[0022] Figure 4 It is a schematic diagram of the pedestrian target trajectory association calculation of the present invention.

[0023] Figure 5 It is a schematic diagram of the pedestrian target trajectory association result of the present invention.

[0024] Figure 6 It is a schematic diagram of pedestrian trajectory prediction and vehicle collision avoidance in the embodiment of the present invention.

[0025] Figure 7 It is a deceleration collision avoidance scenario diagram in the embodiment of the present invention.

[0026] Figure 8 It is a schematic diagram of the deceleration collision avoidance scenario index in the embodiment of the present invention.

[0027] Figure 9 It is a steering collision avoidance scenario diagram in the embodiment of the present invention.

[0028] Figure 10 It is a schematic diagram of the steering collision avoidance scenario index in the embodiment of the present invention. Specific Embodiments

[0029] Embodiment

[0030] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0031] As shown in this application and the claims, unless the context clearly indicates an exception, the words "a", "an", "one" and / or "the" are not specifically singular and may also include plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0032] Unless otherwise specifically stated, the relative arrangements, numerical expressions and values of the components and steps described in these embodiments do not limit the scope of this application. At the same time, it should be understood that for the sake of description, the dimensions of the various parts shown in the drawings are not drawn in actual proportional relationships. Technologies, methods and devices known to those of ordinary skill in the relevant field may not be discussed in detail, but in appropriate cases, the said technologies, methods and devices should be regarded as part of the authorization specification. In all the examples shown and discussed here, any specific value should be construed as merely exemplary and not as a limitation. Therefore, other examples of the exemplary embodiments may have different values. It should be noted that: similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0033] Combined with Figure 1 , a pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning includes the following steps:

[0034] Step 1: Perform pedestrian target detection:

[0035] Step 1-1: Collect the required RGB image data from the CARLA simulation environment or a real camera;

[0036] Step 1-2: Combined with Figure 2 , perform pedestrian detection based on the YOLOv8 network model to obtain the bounding box coordinates (x, y, w, h) and class confidence of the pedestrian target;

[0037] Among them, x and y respectively represent the horizontal and vertical coordinates of the center point of the target detection box, and w and h respectively represent the width and height of the target detection box;

[0038] Among them, combined withFigure 3 The YOLOv8 network model adopts a convolutional block attention mechanism to enhance the detection ability for pedestrians with small targets and pedestrians in complex backgrounds, which is used to enhance the detection ability of YOLOv8 for pedestrians with small targets and pedestrians in complex backgrounds and improve the detection accuracy in occluded environments;

[0039] The convolutional block attention mechanism includes channel attention and spatial attention.

[0040] Step 1-3: Output the pedestrian target detection data in each frame of the image.

[0041] Step 2: Perform pedestrian trajectory tracking based on the pedestrian target detection results:

[0042] Step 2-1: Build a pedestrian trajectory tracking model based on DeepSORT, input the pedestrian target detection box and class information of the pedestrian target detection data output in Step 1, and obtain the pedestrian target trajectory tracking vector:

[0043] x = [u, v, t, h, u, v, r, h] T

[0044] Among them, u and v represent the center coordinates (u, v) of the target detection box, r represents the aspect ratio of the target detection box, h represents the height of the target detection box, and the two sets of parameters respectively represent the motion states of the current frame and the previous frame of the target trajectory;

[0045] For the tracking of bounding boxes by DeepSort, it represents them as a vector x containing eight variables. These variables mainly involve the motion state of the target, and the motion state of the target is analyzed and tracked in an eight-dimensional space. Through this method, DeepSort can effectively perform target tracking and provide more stable performance especially in complex scenarios.

[0046] Step 2-2: Use IoU (Intersection over Union) combined with cosine similarity matching to determine whether the newly detected target trajectory is associated with the existing trajectory, as Figure 4 shown:

[0047]

[0048] Among them, ρ 2 (A center , B center ) represents the distance between the center point of the newly detected target detection box (the target position predicted by the Kalman filter) and the center point of the existing trajectory's target detection box (the target position detected by YOLOv8), that is, the distance of q in Figure 4 and the distance of w A and h ALet \(w\) be the width and height of the newly detected object detection bounding box. B and \(h\) B Let \(w'\) and \(h'\) be the width and height of the object detection bounding box of the existing trajectory, and \(\alpha\) represents the balance coefficient of the matching score, which is used to adjust the influence of different factors (such as IOU and Mahalanobis distance) in the calculation of the association cost. Its value ranges from 0 to 1. Let \(v\) represent the Mahalanobis distance, which represents the matching degree between the target state predicted by the Kalman filter and the detection bounding box. The Mahalanobis distance measures the normalized distance of the detection bounding box relative to the predicted distribution, and the smaller the value, the higher the matching degree.

[0049] The new trajectory can be determined whether it is associated with the existing trajectory through various methods such as comparing the calculated value with the set threshold. The trajectory association result in this embodiment is as Figure 5 shown.

[0050] Step 2-3: Based on the trajectory association result, output the final trajectory tracking result, that is, the trajectory IDs of multiple pedestrians and their historical trajectories.

[0051] Step 3: Perform pedestrian trajectory prediction based on the pedestrian trajectory tracking result:

[0052] Based on the trajectory IDs of pedestrians and their historical trajectories (multi-frame result data), use Social-LSTM to construct a pedestrian trajectory prediction model and output the output result of the pedestrian trajectory prediction model:

[0053] In the model, at each time step, the hidden state of the LSTM represents the current state of the pedestrian, as shown in the figure. Through the social pooling mechanism, these hidden states are shared with other LSTM networks in the neighborhood to form a "social hidden state tensor" Specifically, given a hidden state dimension of \(D\) and a neighborhood size of \(N\) o , as Figure 5 shown; the model constructs an \(N\) o ×\(N\) o ×\(D\) tensor to represent the motion characteristics of neighbors:

[0054]

[0055] where, represents the pedestrian historical trajectory data, 1 mn represents judging whether the neighboring pedestrians are within the range near the target pedestrian, represents the abscissa of the \(j\)-th pedestrian at time \(t\), represents the ordinate of the \(j\)-th pedestrian at time \(t\).

[0056] Step 4: Perform vehicle collision avoidance control based on the pedestrian trajectory prediction data and output the vehicle collision avoidance decision;

[0057] In the CARLA collision avoidance simulation, the results of pedestrian trajectory prediction serve as a crucial input for the vehicle's collision avoidance decision-making, which is used to plan the collision avoidance strategy in advance and dynamically adjust the vehicle's driving state. Specifically, the future trajectories predicted by Social-LSTM are input into the MPC (Model Predictive Control) + PID (Proportional Integral Derivative Control) module to calculate the optimal lateral collision avoidance path and the longitudinal speed adjustment strategy.

[0058] First, the vehicle receives pedestrian information collected by CARLA sensors (such as RGB cameras, LiDAR), and through YOLOv8 + CBAM object detection and DeepSORT trajectory tracking, obtains the current position and historical trajectories of the pedestrians. Then, Social-LSTM predicts the trajectories of pedestrians within the next T_f seconds by modeling the interaction relationships between pedestrians, and combines the current state of the vehicle (position, speed, heading) to calculate whether the pedestrians are likely to enter the vehicle's driving path. As Figure 6 shown.

[0059] Specifically: Based on the pedestrian trajectory prediction results output in Step 3, combined with the current driving state of the vehicle, it is judged whether the pedestrians are likely to enter the vehicle's driving path, and collision avoidance control is performed on the vehicle;

[0060] Among them, using model predictive control, the lateral trajectory of the vehicle is adjusted based on MPC. By solving the optimization problem, an optimal driving trajectory that bypasses the predicted pedestrian path is generated. At the same time, the longitudinal speed of the vehicle is controlled based on PID. If the predicted trajectory indicates that the pedestrians are about to enter the dangerous area, the PID controller reduces the vehicle speed or even brakes to avoid collision;

[0061] The goal of adjusting the vehicle's lateral trajectory based on MPC is to calculate the optimal lateral avoidance trajectory of the vehicle without exceeding the lane limits, ensuring that the predicted pedestrian trajectory is bypassed;

[0062] Using MPC to solve the objective function to obtain the optimal driving trajectory of the vehicle:

[0063]

[0064] Among them, the constraint conditions of the objective function are:

[0065] Control variable constraint, that is, the vehicle steering angle constraint:

[0066] ζ min (k) ≤ ζ(k) ≤ ζ max (k), k = 0, 1, …, t + N C -1

[0067] Output variable constraint, that is, the vehicle's road position constraint, which shall not exceed the road limit range:

[0068] Y min Y(k) ≤ Y(k) ≤ Y max for k = 0, 1, …, t + N C -1

[0069] Centroid sideslip angle constraint:

[0070] The sideslip angle of the vehicle's centroid is crucial for its stability, so it must be restricted. The value range is:

[0071] -arctan(0.02μg) ≤ β ≤ arctan(0.02μg)

[0072] Lateral angular velocity constraint:

[0073] Excessive lateral acceleration may cause discomfort to the vehicle occupants or even loss of control, while too small lateral acceleration may not be able to complete collision avoidance. According to the analysis, the lateral acceleration needs to be restricted within the following range:

[0074] -0.35g ≤ a y ≤ 0.35g

[0075] Tire sideslip angle constraint

[0076] For the sideslip angle of the tire, the front - wheel sideslip angle should be restricted within the following range to ensure high fitting accuracy of the tire:

[0077] -3° ≤ α f, ≤ 3°

[0078] where e y represents the lateral error of the vehicle deviating from the target path, e θ represents the heading - angle error of the vehicle, δ represents the steering - wheel angle change rate, to avoid sharp steering, J is the MPC control objective function, w1 represents the lateral - error weight, which is used to control the influence of the lateral error of the vehicle deviating from the target path on the optimization objective, w2 represents the heading - angle error weight, which is used to adjust the influence of the heading - angle error in the optimization objective, w3 represents the steering - wheel angle change - rate weight, which is used to restrict the steering - wheel angle change rate, avoid sudden sharp steering of the vehicle, and improve the driving smoothness, ζ(k) represents the vehicle steering - angle constraint, N C represents the end time of the collision - avoidance simulation run, Y(k) represents the vehicle's road - position constraint, μ represents the road friction coefficient, β represents the sideslip angle of the vehicle's centroid, a y represents the vehicle's lateral angular velocity, α f represents the vehicle's tire sideslip angle.

[0079] The goal of controlling the vehicle's longitudinal speed based on PID is to control the vehicle's acceleration / deceleration to ensure stopping or continuing to drive within a safe range;

[0080] In the longitudinal collision avoidance control model, it is first necessary to plan the optimal deceleration to ensure that the vehicle decelerates safely within the collision avoidance area and ensure passenger comfort. When the vehicle decelerates, it only needs to ensure that the deceleration distance is less than the initial longitudinal relative distance ΔS′ within the pedestrian crossing time tcross, that is, to ensure that there is no collision risk. That is:

[0081]

[0082] First, plan the optimal deceleration a d and the speed v e after braking deceleration to ensure that the vehicle decelerates safely within the collision avoidance area and ensure passenger comfort:

[0083]

[0084] v e = v v + a d t cross

[0085] where ζ represents the distance elasticity factor, t cross represents the pedestrian crossing time, ΔS′ represents the initial longitudinal relative distance, and v v represents the initial vehicle motion speed.

[0086] In addition, when the vehicle makes an emergency braking collision avoidance, the vehicle needs to perform full braking to ensure a complete stop to avoid collision. In this case, the braking deceleration must meet the following conditions to ensure the safety of people and vehicles:

[0087]

[0088] At this time, the optimal deceleration during the vehicle's full braking can be obtained as:

[0089]

[0090] In this embodiment, the standard situation of crossing the road is used to conduct the collision avoidance control experiment:

[0091] Experimental objective: To test the collision avoidance ability of the vehicle collision avoidance system in the scenario of conventional pedestrian crossing the road.

[0092] Experimental steps:

[0093] (1) Pedestrian detection: In the CARLA simulation environment, pedestrians start crossing the road from the roadside at a speed of 1.2 m / s, and YOLOv8 identifies the pedestrian position and marks the detection frame.

[0094] (2) Trajectory tracking: DeepSORT tracks the pedestrians, generates historical trajectories, and provides input for the prediction module.

[0095] (3) Trajectory prediction: Social-LSTM predicts the future 3s trajectory of pedestrians and shows the possible positions where pedestrians may reach.

[0096] (4) Collision avoidance control:

[0097] MPC calculates the lateral avoidance path, determines the optimal driving trajectory, and causes the vehicle to deviate appropriately to bypass the pedestrian walking route.

[0098] PID longitudinal control calculates the target speed of the vehicle. If the pedestrian is too close, it brakes to 0m / s to ensure safety.

[0099] The experimental results are as Figure 7 and Figure 8 shown:

[0100] Among them Figure 7 (a) to Figure 7 (d) are respectively the deceleration collision avoidance scenario, the pedestrian starts to cross the street, deceleration collision avoidance, and the pedestrian passes through the dangerous area;

[0101] Figure 8 (a) to Figure 8 (d) are respectively the schematic diagrams of the vehicle's speed - acceleration - braking force percentage, speed - TTC, speed - tcross, longitudinal and lateral distances between the vehicle and the pedestrian;

[0102] In this embodiment, the pedestrian detection accuracy mAP@0.5 = 89.6%, and the target can still be detected under occlusion.

[0103] The trajectory prediction error (FDE) is reduced by 17.5%, and the predicted trajectory is closer to the true trajectory.

[0104] The collision avoidance success rate is 98%. The vehicle adjusts the trajectory in advance, successfully avoids the pedestrian, and no collision occurs.

[0105] In this embodiment, a collision avoidance control experiment is also carried out with sudden pedestrian intrusion:

[0106] Experimental objective: To test the response ability of the collision avoidance system when a pedestrian suddenly rushes out.

[0107] Experimental steps:

[0108] Scenario setting: A pedestrian suddenly rushes into the vehicle's driving route at a speed of 3m / s.

[0109] YOLOv8 identifies the target, DeepSORT performs trajectory tracking, and Social-LSTM predicts the next movement trend of the pedestrian.

[0110] Collision avoidance control:

[0111] MPC plans a new lateral trajectory and judges whether there is enough space for lane deviation.

[0112] The PID control performs emergency braking, calculates the safe stopping distance, and controls the braking force.

[0113] The experimental results are as Figure 9 and Figure 10 shown:

[0114] Among them Figure 9 (a) to Figure 9 (d) are respectively the emergency steering collision avoidance scenario, the pedestrian starts to cross the street, the emergency steering collision avoidance, and the vehicle stops;

[0115] Figure 10 (a) to Figure 10 (d) are respectively the schematic diagrams of the vehicle's speed - acceleration - braking force percentage, speed - TTC, speed - tcross, and the longitudinal and lateral distances between the vehicle and the pedestrian;

[0116] In this embodiment, the vehicle makes a braking decision within 0.8 s to avoid collision.

[0117] The error of the vehicle's braking stop distance is controlled within ±0.2 m, meeting the safe stopping standard.

[0118] The false alarm rate is less than 5%, and the system will not falsely trigger the brakes for the normal movement of pedestrians.

[0119] In summary, the present invention has been tested in the CARLA simulation environment through YOLOv8 + CBAM object detection, DeepSORT trajectory tracking, Social - LSTM trajectory prediction, and MPC + PID collision avoidance control, verifying the effectiveness and safety of the system in complex traffic environments. The experimental results show that this method can effectively reduce the trajectory prediction error, improve the collision avoidance success rate, and has wide application value in the field of autonomous driving.

[0120] The above - described embodiments only represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning, characterized in that: The following steps are involved: Step 1: Perform pedestrian target detection; Step 2: Pedestrian trajectory tracking based on pedestrian target detection results; Step 3: Pedestrian trajectory prediction based on the pedestrian trajectory tracking result; Step 4: Perform vehicle collision avoidance control based on pedestrian trajectory prediction data and output vehicle collision avoidance decision.

2. The pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning according to claim 1 is characterized in that: The pedestrian target detection in step 1 is specifically as follows: Step 1-1, collect the required RGB image data; Step 1-2: Perform pedestrian detection based on the YOLOv8 network model to obtain the bounding box coordinates (x, y, w, h) and category confidence of the pedestrian target; Where x and y represent the horizontal and vertical coordinates of the center point of the target detection frame, respectively, and w and h represent the width and height of the target detection frame, respectively; Step 1-3: output the pedestrian target detection data detected in each frame of the image.

3. The pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning according to claim 2 is characterized in that: The YOLOv8 network model adopts the convolutional block attention mechanism to enhance the detection capability of small target pedestrians and pedestrians in complex backgrounds; The convolution block attention mechanism includes channel attention and spatial attention.

4. The pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning according to claim 2 is characterized in that: The pedestrian trajectory tracking in step 2 is specifically as follows: Step 2-1: Build a pedestrian trajectory tracking model based on DeepSORT, input the pedestrian target detection frame and category information of the pedestrian target detection data output in step 1, and obtain the pedestrian target trajectory tracking vector: x=[u,v,r,h,u,v,r,h] T Among them, u, v represent the center coordinates (u, v) of the target detection frame, r represents the aspect ratio of the target detection frame, h represents the height of the target detection frame, and the two sets of parameters represent the motion states of the current frame and the previous frame of the target trajectory respectively; Step 2-2: Use IoU combined with cosine similarity matching to determine whether the newly detected target trajectory is associated with the existing trajectory: Among them, ρ 2 (A center ,B center ) represents the distance between the center point of the newly detected target detection box and the center point of the target detection box of the existing trajectory, w A and h A is the width and height of the newly detected target detection box, w B and h B is the width and height of the target detection box of the existing trajectory, α represents the balance coefficient of the matching score, and v represents the Mahalanobis distance; Step 2-3: Based on the trajectory association result, the final trajectory tracking result is output, that is, the trajectory IDs of multiple pedestrians and their historical trajectories.

5. The pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning according to claim 4 is characterized in that: The pedestrian trajectory prediction in step 3 is specifically as follows: Based on the pedestrian’s trajectory ID and historical trajectory, Social-LSTM is used to build a pedestrian trajectory prediction model, and the pedestrian trajectory prediction model output results are output: in, Represents pedestrian historical trajectory data, 1 mn Indicates whether the neighboring pedestrians are within the vicinity of the target pedestrian. represents the horizontal coordinate of the jth pedestrian at time t, represents the ordinate of the j-th pedestrian at time t.

6. The pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning according to claim 1, characterized in that: The vehicle collision avoidance control in step 4 is specifically as follows: Based on the pedestrian trajectory prediction result output in step 3 and combined with the current state of the vehicle, determine whether the pedestrian is likely to enter the vehicle's driving path and perform collision avoidance control on the vehicle; Among them, model predictive control is used to adjust the vehicle's lateral trajectory based on MPC. By solving the optimization problem, an optimal driving trajectory that bypasses the predicted path of pedestrians is generated. At the same time, the vehicle's longitudinal speed is controlled based on PID. If the predicted trajectory indicates that the pedestrian is about to enter the dangerous area, the PID controller reduces the vehicle speed or even stops it to avoid a collision.

7. The pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning according to claim 6 is characterized in that: The MPC-based adjustment of the vehicle lateral trajectory is specifically as follows: Use MPC to solve the objective function and obtain the optimal driving trajectory of the vehicle: Among them, the constraints of the objective function are: ζ min (k)≤ζ(k)≤ζ max (k),k=0,1,…,t+N C -1 Y min (k)≤Y(k)≤Y max (k),k=0,1,…,t+N C -1 -arctan(0.02μg)≤β≤arctan(0.02μg) <h2 style=";text-align:left;direction:ltr">-0.35g≤a<h2 style=";text-align:left;direction:ltr"> y <h2 style=";text-align:left;direction:ltr"> ≤0.35g -3°≤α f ≤3° Among them, e y represents the lateral error of the vehicle from the target path, e θ represents the vehicle heading angle error, δ represents the steering wheel angle change rate to avoid drastic steering, J is the MPC control objective function, w1 represents the lateral error weight, w2 represents the heading angle error weight, w3 represents the steering wheel angle change rate weight, ζ(k) represents the vehicle steering angle constraint, N C represents the end time of collision avoidance simulation, Y(k) represents the road position constraint of the vehicle, μ represents the road friction coefficient, β represents the side panel angle of the vehicle center of mass, and a y represents the vehicle's lateral angular velocity, α f Indicates the side angle of the vehicle tire.

8. The pedestrian trajectory prediction and intelligent collision avoidance control method based on deep learning according to claim 6 is characterized in that: The PID-based control of the vehicle longitudinal speed is specifically as follows: First, plan the optimal deceleration a when the vehicle decelerates d and the speed v after braking e To ensure safe deceleration of the vehicle in the collision avoidance zone and ensure passenger comfort: v e =v v +a d t cross Where ζ represents the distance elasticity factor, t cross represents the pedestrian crossing time, ΔS′ represents the initial longitudinal relative distance, v v Indicates the initial speed of the vehicle. Among them, when the vehicle brakes urgently to avoid collision, the vehicle needs to brake with full force to ensure complete stop to avoid collision. The optimal deceleration at this time is: 。

Images