Method and device for predicting turning intention

Abstract

This specification discloses a method and apparatus for predicting turning intentions. An unmanned driving device acquires an image of an obstacle and extracts a directional gradient histogram of the obstacle's boundary. Based on the directional gradient histogram and a trained first classifier, the device determines the obstacle's first turning intention. The device then uses multiple trained second classifiers to determine the obstacle's second turning intentions in each possible direction of travel. Finally, based on the first and second turning intentions, the device determines the obstacle's final turning intention, thereby determining an obstacle avoidance strategy based on the final turning intention. By predicting the obstacle's turning intention based on the directional gradient histogram of the obstacle's boundary, computational complexity is reduced, prediction timeliness is improved, and prediction accuracy is balanced by using multiple classifiers to determine the final turning intention.

Description

Technical Field

[0001] This specification relates to the field of autonomous driving, and in particular to a method and device for predicting steering intention. Background Art

[0002] With the development of autonomous driving technology, the safety of unmanned vehicles on the road has attracted increasing attention. To ensure the safety of unmanned vehicles on the road, it is necessary to predict the turning intentions of dynamic obstacles on the road to control the unmanned vehicles to avoid obstacles.

[0003] In the prior art, when predicting the turning intention of a dynamic obstacle traveling on the road, it is necessary to obtain characteristic information such as the speed, size, and position of the dynamic obstacle, and then predict the trajectory of the dynamic obstacle based on this characteristic information. Finally, based on the predicted trajectory of the dynamic obstacle, the turning intention of the dynamic obstacle is predicted.

[0004] However, the algorithms for predicting the trajectory of dynamic obstacles are complex, requiring high computing power from the autonomous vehicle and taking a long time to generate, resulting in low timeliness of prediction results. This makes it difficult for autonomous vehicles to predict the turning intentions of dynamic obstacles and avoid them in advance, making it difficult to ensure the safety of autonomous vehicles. Summary of the Invention

[0005] This specification provides a method and device for predicting turning intention to partially solve the above-mentioned problems existing in the prior art.

[0006] This manual adopts the following technical solutions:

[0007] This specification provides a method for predicting turning intention, including:

[0008] Obtain an image of the obstacle;

[0009] Extracting a directional gradient histogram representing the boundary of the obstacle according to the image;

[0010] Inputting the directional gradient histogram into a trained first classifier to obtain a first turning intention of the obstacle output by the first classifier;

[0011] Inputting the directional gradient histogram of the obstacle into each second classifier corresponding to each trained turning intention, respectively, to obtain each second turning intention of the obstacle as output by each second classifier;

[0012] A final turning intention of the obstacle is determined according to the first turning intention and the second turning intentions, so as to determine an obstacle avoidance strategy based on the final turning intention.

[0013] Optionally, before extracting the directional gradient histogram representing the obstacle boundary from the image, the method further includes:

[0014] According to the image, performing grayscale processing on the image;

[0015] According to the grayscale-processed image and the preset size, the image is converted into an image with a preset size.

[0016] Optionally, extracting a directional gradient histogram representing the obstacle boundary based on the image specifically includes:

[0017] According to the image, dividing the image into a plurality of regions;

[0018] For each region, according to the preset gradient directions, determine the eigenvectors corresponding to the gradient directions in the region;

[0019] A directional gradient histogram representing the boundary of the obstacle is determined according to the feature vector corresponding to each gradient direction in each area.

[0020] Optionally, inputting the directional gradient histogram into a trained first classifier to obtain a first turning intention of the obstacle output by the first classifier specifically includes:

[0021] According to the directional gradient histogram and the trained first classifier, the directional gradient histogram is used as the input of the first classifier, and the output result of the first classifier is obtained as the first turning intention of the obstacle.

[0022] Optionally, the directional gradient histogram of the obstacle is input into each trained second classifier corresponding to each turning intention, and each second turning intention of the obstacle output by each second classifier is obtained, specifically including:

[0023] According to the directional gradient histogram, as well as the trained right-turn second classifier, straight-ahead second classifier, left-turn second classifier, and U-turn second classifier, the directional gradient histogram is used as input to determine the output of the right-turn second classifier as the right-turn second turning intention of the obstacle, the output of the straight-ahead second classifier as the straight-ahead second turning intention of the obstacle, the output of the left-turn second classifier as the left-turn second turning intention of the obstacle, and the output of the U-turn second classifier as the U-turn second turning intention of the obstacle.

[0024] Optionally, determining a final turning intention of the obstacle according to the first turning intention and each of the second turning intentions specifically includes:

[0025] determining, according to the first turning intention, a first confidence level of a direction corresponding to the first turning intention;

[0026] determining, according to each second turning intention, a second confidence level for each direction corresponding to each second turning intention;

[0027] A final turning intention of the obstacle is determined according to a first confidence level of the direction corresponding to the first turning intention and a second confidence level of each direction corresponding to each second turning intention.

[0028] Optionally, the method further includes:

[0029] determining a gradient direction according to the direction corresponding to the first turning intention, and determining a reference confidence level according to a first confidence level corresponding to the first turning intention;

[0030] For each direction corresponding to the second turning intention, determining a number of steps for transforming the direction into the gradient direction, and determining a gradient between the direction corresponding to the second turning intention and the gradient direction according to the number of steps;

[0031] A second confidence level for the direction corresponding to each second turning intention is determined according to each second turning intention, the direction corresponding to each second turning intention, the gradient of the gradient direction, and the reference confidence level.

[0032] This specification provides a device for predicting turning intention, comprising:

[0033] Acquisition module: used to acquire images of obstacles;

[0034] A characterization module is configured to extract a directional gradient histogram representing the boundary of the obstacle based on the image;

[0035] The first module is configured to input the directional gradient histogram into a trained first classifier to obtain a first turning intention of the obstacle output by the first classifier;

[0036] The second module is configured to input the directional gradient histogram of the obstacle into each second classifier corresponding to each trained turning intention, and obtain each second turning intention of the obstacle as output by each second classifier;

[0037] A determination module is configured to determine a final turning intention of the obstacle according to the first turning intention and each of the second turning intentions, so as to determine an obstacle avoidance strategy based on the final turning intention.

[0038] This specification provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the method for predicting turning intention is implemented.

[0039] This specification provides an unmanned driving device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for predicting steering intention when executing the program.

[0040] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:

[0041] In the method for predicting turning intention provided in this specification, the unmanned driving device can obtain an image of an obstacle and extract a directional gradient histogram of the image to determine the first turning intention of the obstacle based on the directional gradient histogram and a trained first classifier, and use multiple second classifiers to determine the second turning intentions of the obstacle for each direction in which the obstacle may travel. Finally, based on the first turning intention and each second turning intention, the final turning intention of the obstacle is determined to determine the obstacle avoidance strategy based on the final turning intention.

[0042] It can be seen from the above method that this method predicts the turning intention of the obstacle by using the directional gradient histogram of the obstacle boundary, which reduces the computational complexity, improves the prediction timeliness, and takes into account the prediction accuracy by determining the final turning intention through multiple classifiers. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The drawings described herein are used to provide a further understanding of this specification and constitute a part of this specification. The exemplary embodiments and descriptions of this specification are used to explain this specification and do not constitute an improper limitation of this specification. In the drawings:

[0044] Figure 1 A flowchart of a method for predicting turning intention in this specification;

[0045] Figure 2 A schematic diagram of an obstacle steering provided in this manual;

[0046] Figure 3 A schematic diagram of a structure for predicting turning intention provided in this manual;

[0047] Figure 4 A schematic diagram of a device for predicting turning intention provided in this specification;

[0048] Figure 5 The corresponding Figure 1 Schematic diagram of unmanned driving equipment. DETAILED DESCRIPTION

[0049] To make the purpose, technical solutions, and advantages of this specification more clear, the technical solutions of this specification will be clearly and completely described below in conjunction with the specific embodiments of this specification and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0050] The technical solutions provided by the embodiments of this specification are described in detail below with reference to the accompanying drawings.

[0051] Figure 1 The following is a flowchart of predicting turning intention in this specification, which specifically includes the following steps:

[0052] S100: Obtain an image of the obstacle.

[0053] Because traffic participants typically only perform turns at intersections, the method for predicting turning intentions provided herein can be used to predict the turning intentions of obstacles at intersections. Furthermore, because intersection traffic conditions are complex and rapidly changing, and because predicting the turning intentions of obstacles is intended for obstacle avoidance by unmanned vehicles, the timeliness of the prediction is highly critical. Therefore, in one or more embodiments of this specification, the method for predicting turning intentions can be performed by unmanned vehicles.

[0054] Specifically, first, when the unmanned driving device drives to an intersection, the image sensor on the unmanned driving device can collect images containing various obstacles in the intersection.

[0055] Among them, the unmanned driving device can determine whether it has reached the intersection based on the positioning data received by the chip in the Global Positioning System (GPS), and obtain the image of the intersection collected by the image sensor when it is determined to have arrived. Of course, in this manual, the unmanned driving device can also use other methods to determine whether it has reached the intersection. For example, the target object recognition method is used to determine whether there is a traffic light within a preset distance in front of the unmanned driving device. If so, it is determined that the intersection has been reached, and so on. The specific method used to determine whether the unmanned driving device has reached the intersection can be set as needed, and this manual does not impose any restrictions.

[0056] Secondly, since the image may contain obstacles such as motor vehicles, non-motor vehicles, pedestrians, trees, and buildings, and only dynamic obstacles require steering intention prediction, the unmanned driving device can perform obstacle recognition on the collected image and identify dynamic obstacles as obstacles in the image. Dynamic obstacles include motor vehicles, non-motor vehicles, pedestrians, and other obstacles that participate in traffic.

[0057] Of course, how to identify obstacles and determine the categories of obstacles is already a relatively mature technology, and this manual does not limit this, and it can be set as needed.

[0058] Finally, based on the identified obstacle in the image, the area corresponding to the obstacle in the image is determined, and the image of the area is extracted as the image of the obstacle.

[0059] Of course, after the target object recognition is performed on the image collected by the unmanned driving device, there may be one or more obstacles in the image. Therefore, for each obstacle contained in the image, this manual can be used to identify the obstacle. Figure 1 The provided method determines the turning intention of the obstacle. For the convenience of description, this manual uses the example of an image captured by the unmanned driving device containing only one obstacle after target object recognition.

[0060] The unmanned driving equipment mentioned in this specification may refer to autonomous vehicles, robots, automatic delivery equipment, and other devices capable of autonomous driving. Therefore, unmanned driving equipment that utilizes the method for predicting steering intentions provided in this specification can be used to perform driving tasks in the unmanned driving field and delivery tasks in the delivery field. For example, unmanned driving equipment can be used in business scenarios such as express delivery, logistics, and food delivery.

[0061] S102: Extracting a directional gradient histogram representing the obstacle boundary based on the image.

[0062] Since the method of predicting the trajectory of obstacles based on neural networks to determine the obstacle avoidance strategy requires a lot of computing power for trajectory prediction. However, the computing power of the unmanned driving device is limited, and determining the obstacle avoidance strategy also requires a certain amount of computing power. In order to avoid the problem of high computing power and resource consumption in trajectory prediction. In one or more embodiments of this specification, the unmanned driving device can use a histogram of oriented gradient (HOG) to determine the boundary of the obstacle, and then classify the histogram of oriented gradient through multiple second classifiers to predict the turning intention of the obstacle for obstacle avoidance. Since it is relatively simple to determine the HOG algorithm, and the classifier is generally simpler than the neural network model, it can effectively avoid resource occupation and reduce prediction time.

[0063] Specifically, the unmanned driving device may first resize the extracted image of the area to an image of a preset size based on a preset size, so as to facilitate determining the HOG features representing the boundary of the obstacle. Based on the resized image, the image is then divided into a number of regions. For each region of the image, the gradient magnitude and gradient direction of each pixel in the region are determined. Then, based on the preset gradient directions, the gradient directions, and gradient magnitudes of each pixel in the region, the gradient magnitude of each pixel along the preset gradient directions is determined.

[0064] Next, for each preset gradient direction, the gradient magnitudes of each pixel in the region along that gradient direction are accumulated to determine the feature vector for that gradient direction within the region. Based on the feature vectors corresponding to each gradient direction within each region of the image, the HOG features representing the obstacle boundary are determined.

[0065] Of course, in one or more embodiments of the specification, other methods can be used to represent the boundary information of the obstacle, for example, the boundary information of the obstacle can be determined based on image semantic segmentation technology. Of course, other methods can also be used to represent the boundary information of the obstacle, which can be set as needed and is not limited in this specification.

[0066] By determining the HOG features that characterize the boundary of the obstacle, the information of the obstacle in the intersection can be quickly determined while occupying less computing power, thereby predicting the turning intention of the obstacle in the intersection.

[0067] S104: Inputting the directional gradient histogram into a trained first classifier to obtain a first turning intention of the obstacle output by the first classifier.

[0068] In one or more embodiments of the present specification, after determining the HOG features that characterize the boundary of the obstacle, the turning intention of the obstacle can be predicted based on the HOG features using a trained classifier, thereby quickly determining the predicted result of the turning intention of the obstacle while occupying less computing power, and determining an obstacle avoidance strategy based on the predicted result.

[0069] Specifically, the unmanned driving device can use the determined HOG features representing the boundary of the obstacle and the trained first classifier as the input of the first classifier to obtain the output result of the first classifier as the first turning intention of the obstacle.

[0070] The first classifier may be a support vector machine (SVM) classifier. Of course, other classifiers may also be used, such as a random forest classifier, a naive Bayes classifier, etc. The specific classifier used may be set as needed and is not limited in this specification.

[0071] For the convenience of description, the obstacle is taken as a motor vehicle. Figure 2 As shown, Figure 2 This is a schematic diagram of an obstacle steering method provided in this specification. While driving through an intersection, the unmanned driving device 200 captures an image of a motor vehicle 202 located at the intersection and determines that the motor vehicle 202 can perform one of four possible directions at the intersection: right turn, straight ahead, left turn, or U-turn. As shown in the figure, the lines indicating the direction of travel of the motor vehicle 202, from left to right, represent right turn, straight ahead, left turn, and U-turn, respectively. Therefore, the first steering intention can be one of the four directions of travel of the obstacle: right turn, straight ahead, left turn, or U-turn.

[0072] When training the first classifier, the same process as steps S100 to S102 can be used to determine the HOG features of each obstacle contained in each image based on images containing obstacles obtained by any means. The determined HOG features of each obstacle are used as training samples, and each training sample is annotated using a manual annotation method. The first classifier is then trained using each annotated training sample.

[0073] By adopting the above method, the HOG features that characterize the boundaries of obstacles in the intersection can be processed by the first classifier to determine the output result of the first classifier while occupying less computing power, thereby predicting the turning intention of obstacles in the intersection and promptly giving the prediction results of the turning intention of obstacles in the intersection, so that the unmanned driving device can determine the obstacle avoidance strategy based on the prediction results.

[0074] S106: Inputting the directional gradient histogram of the obstacle into each trained second classifier corresponding to each turning intention, respectively, to obtain each second turning intention of the obstacle as output by each second classifier.

[0075] In one or more embodiments of this specification, in order to more accurately predict the turning intention of an obstacle at an intersection, after determining the first turning intention of the obstacle output by the first classifier, the HOG features representing the obstacle boundary may be input into multiple trained second classifiers to determine the second turning intentions of the obstacle output by each of the multiple second classifiers. Based on the first turning intention and the multiple second turning intentions, the final turning intention of the obstacle is determined, thereby achieving a more accurate prediction of the turning intention of the obstacle at the intersection. Each second classifier predicts the turning intention in each direction.

[0076] Specifically, the unmanned driving device can use the determined HOG feature representing the boundary of the obstacle, as well as the trained right-turn second classifier, straight-ahead second classifier, left-turn second classifier, and U-turn second classifier, as the input of the right-turn second classifier, determine the output of the right-turn second classifier as the right-turn second turning intention of the obstacle, use the HOG feature as the input of the straight-ahead second classifier, determine the output of the straight-ahead second classifier as the straight-ahead second turning intention of the obstacle, use the HOG feature as the input of the left-turn second classifier, determine the output of the left-turn second classifier as the left-turn second turning intention of the obstacle, and use the HOG feature as the input of the U-turn second classifier, determine the output of the U-turn second classifier as the U-turn second turning intention of the obstacle.

[0077] Among them, the second right-turn classifier, the second straight-ahead classifier, the second left-turn classifier, and the second U-turn classifier are all trained second classifiers, and each of the above second classifiers can only output the turning intention of the obstacle in the direction corresponding to the second classifier. For example, the second right-turn classifier can only output yes or no. Yes means that the turning intention of the obstacle is to turn right, and no means that the turning intention of the obstacle is not to turn right. Of course, yes or no is only one form of output, and the second right-turn classifier can also output data in other forms, which can be set as needed, and this manual does not limit it.

[0078] It should be noted that when training the second classifier, the image containing obstacles can be determined by any method, and the method in this manual can be used. Figure 1 The provided method determines the HOG features of obstacles contained in each image, uses the determined HOG features of each obstacle as training samples, and uses a manual labeling method to label each training sample. This allows each labeled training sample to be used to train each second classifier.

[0079] Using the above method, the unmanned driving device can determine multiple second turning intentions of the obstacle, so that the unmanned driving device can determine the final turning intention of the obstacle based on the first turning intention and multiple second turning intentions, thereby achieving a more accurate prediction of the turning intention of the obstacle.

[0080] S108: Determine a final turning intention of the obstacle according to the first turning intention and the second turning intentions, and determine an obstacle avoidance strategy based on the final turning intention.

[0081] In one or more embodiments of this specification, after determining the first turning intention and each second turning intention, the final turning intention of the obstacle can be determined based on the first turning intention and each second turning intention, thereby realizing the prediction of the turning intention of the obstacle, so as to determine the obstacle avoidance strategy based on the prediction result.

[0082] like Figure 3 As shown, Figure 3 This is a schematic diagram of a structure for predicting turning intentions provided in this specification. The unmanned driving device can extract HOG features from an image and input them into a first classifier and each second classifier. The output of the first classifier is determined as a first turning intention, and the output of each second classifier is determined as multiple second turning intentions. Based on the first turning intention and the multiple second turning intentions, a final turning intention is determined. This enables prediction of obstacle turning intentions and thus determines an obstacle avoidance strategy.

[0083] Specifically, first, the unmanned driving device may determine the direction corresponding to the first turning intention based on the determined first turning intention of the obstacle, and determine a first confidence level for the direction based on a preset first value, while determining the first confidence levels of other directions to be 0. For example, if the direction of the first turning intention is a left turn and the preset first value is 1, the first confidence level for the left turn is 1, the first confidence level for the right turn is 0, the first confidence level for going straight is 0, and the first confidence level for a U-turn is 0.

[0084] Next, based on the first and second confidence levels for each direction, the confidence levels for the same direction are summed to determine the intention confidence level for each direction. For example, if the first confidence level for a left turn is 1 and the second confidence level for a left turn is 0.6, the left turn intention confidence level is 1.6. The intention confidence levels for each direction are then used as the final turning intention for the obstacle. For example, the final turning intention is (right turn: 0.7, straight ahead: 0.8, left turn: 0.5, U-turn: 0).

[0085] based on Figure 1In the method for predicting turning intention shown in FIG, the unmanned driving device can obtain an image of an obstacle and extract HOG features of the image to determine the first turning intention of the obstacle based on the HOG features and the trained first classifier, and use multiple binary classifiers to determine the second turning intentions of the obstacle for each possible direction of travel of the obstacle. Finally, based on the first turning intention and the second turning intentions, the final turning intention of the obstacle is determined to determine the obstacle avoidance strategy based on the final turning intention.

[0086] It can be seen from the above method that this method predicts the turning intention of the obstacle by using the directional gradient histogram of the obstacle boundary, which reduces the computational complexity, improves the prediction timeliness, and takes into account the prediction accuracy by determining the final turning intention through multiple classifiers.

[0087] In addition, in one or more embodiments of this specification, in order to take into account the Figure 1 The computing power requirements of the provided method and the timeliness of determining the obstacle avoidance strategy of the unmanned driving device can be implemented using other devices. Figure 1 The method provided, for example, the computing platform, the computing terminal that is closer to the unmanned driving device in the network topology, etc., does not limit the specific device used in this specification and can be set as needed.

[0088] In addition, in one or more embodiments of the present specification, in order to reduce the influence of factors such as lighting and color in the environment on the HOG features of the image, the unmanned driving device may pre-process the image before step S102.

[0089] Specifically, the image is grayscaled. Of course, other methods can also be used to pre-process the image, such as image filtering algorithms, gamma correction, etc. The specific method used to pre-process the image can be set as needed and is not limited in this specification.

[0090] By adopting the above method, the unmanned driving device can reduce the influence of factors such as lighting and color in the environment on the image, making it easier to extract the HOG features of the image.

[0091] In addition, in one or more embodiments of this specification, this specification Figure 1 The provided method predicts an obstacle's turning intention once. However, due to the complex road conditions at intersections and the varying movements of obstacles, the autonomous driving system can also predict the turning intention of obstacles within the intersection multiple times. For example, predictions can be made every 10 milliseconds, every meter traveled, and so on. This can be configured as needed and is not limited in this manual.

[0092] Moreover, in the process of predicting the turning intention of each obstacle in the intersection, the obstacle avoidance strategy of the unmanned driving device can be determined based on the prediction results of the turning intention of each obstacle.

[0093] Furthermore, the unmanned driving device may update the obstacle avoidance strategy when the direction with the highest confidence changes in two consecutive final turning intentions. For example, in the previous prediction result, the direction with the highest confidence is a left turn, and in this prediction result, the direction with the highest confidence is a right turn, the obstacle avoidance strategy is updated. Alternatively, in two consecutive final turning intentions, when the change in the difference in confidence of each direction is greater than a preset threshold, the obstacle avoidance strategy may be updated. For example, in two consecutive prediction results, the difference between the highest confidence value and the lowest confidence value is 1 and 0.1 respectively, and the preset threshold is 0.5. Since 1-0.1=0.9>0.5, the change is greater than the threshold, and the obstacle avoidance strategy is updated. Of course, other methods may also be used to determine whether to update the obstacle avoidance strategy. The specific method of determining whether to update the obstacle avoidance strategy can be set as needed, and this manual does not impose any restrictions.

[0094] In addition, in one or more embodiments of the present specification, in order to more accurately extract the HOG features representing the boundary of the obstacle, in step S102 , the HOG features representing the boundary of the obstacle may be extracted in the following manner.

[0095] Specifically, first, the unmanned driving device can adjust the extracted image of the area to an image of a preset size according to a preset size, so as to determine the HOG features that characterize the boundary of the obstacle. And based on the adjusted image, the image is divided into several areas. And for each area, a horizontal filter and a vertical filter are used to calculate the gradient amplitude of each pixel in the area in the horizontal direction, and the gradient amplitude of each pixel in the area in the vertical direction, and then determine the gradient amplitude of each pixel in the area and the gradient direction of each pixel in the area. Among them, the horizontal filter and the vertical filter used can be set as needed. For example, using (1, 0, -1) and (1, 0, -1) T Two filters are used for calculation, or a Sobel filter of size 3*3 is used for calculation. Of course, other filters and filters of other sizes can also be used. They can be set as needed and are not limited in this manual.

[0096] The above process of determining the gradient magnitude and gradient direction of each pixel can be expressed by the formula:

[0097]

[0098] θ=arctan(DY / DX)

[0099] Wherein, for each pixel, D is the gradient magnitude of the pixel, DX is the horizontal gradient magnitude of the pixel, DY is the vertical gradient magnitude of the pixel, and θ is the gradient direction of the pixel. Specifically, the gradient magnitude of the pixel is determined based on the horizontal and vertical gradient magnitudes of the pixel, and the arctan function is used to determine the gradient direction of the pixel based on the horizontal and vertical gradient magnitudes of the pixel.

[0100] Secondly, according to the preset gradient directions, the gradient directions of the pixels in the region, and the gradient amplitudes of the pixels, the gradient amplitudes of the pixels in the gradient directions are determined.

[0101] Again, for each gradient direction, according to the gradient amplitude of each pixel point in the gradient direction in the region, the gradient amplitude of each pixel point in the gradient direction is accumulated to determine the gradient histogram of the gradient direction in the region, and then the gradient histogram of each gradient direction in the region is determined.

[0102] Then, according to the gradient histogram corresponding to each gradient direction in the region, the gradient amplitude in each gradient direction in the region is obtained, and the sum of the gradient amplitudes in each gradient direction in the region is determined as the total gradient value of the region.

[0103] Finally, based on the total gradient value of the region and the gradient magnitudes in each gradient direction within the region, the proportion of the gradient magnitudes in each gradient direction within the region to the total gradient value within the region is determined, thus normalizing the gradient magnitudes in each gradient direction within the region. Furthermore, based on the proportion of the normalized gradient magnitudes in each gradient direction within the region to the total gradient value within the region, a HOG feature map representing the boundary of the obstacle is determined.

[0104] In addition, in one or more embodiments of this specification, in order to save time in labeling training samples, in steps S104 and S106, when training the first classifier and each second classifier, when obtaining training samples, a high-precision map, a global positioning system, or the like can be used to directly obtain an image containing information indicating the lane where the obstacle is located. For example, if there is only one motor vehicle in the image, located at the exit of the left-turn lane of the intersection, it can be determined that the motor vehicle's turning intention is to turn right. Of course, other methods can also be used to determine the turning intention of the obstacle before extracting the HOG features of the obstacle. The specific method can be set as needed and is not limited by this specification.

[0105] Furthermore, in one or more embodiments of this specification, to more accurately determine the obstacle's turning intention, in step S104, when determining the first turning intention, the first turning intention may be the confidence levels of four driving scenarios for the obstacle. Specifically, the confidence levels of the four driving scenarios: right turn, straight ahead, left turn, and U-turn. For example, the first turning intention may be (right turn: 0.6, straight ahead: 0.3, left turn: 0.8, U-turn: 0.2).

[0106] Furthermore, in one or more embodiments of the present specification, in order to more accurately determine the turning intention of the obstacle, in step S108, when determining the second confidence level corresponding to each direction, the following method may be used for determination.

[0107] Specifically, the unmanned driving device can first determine the first confidence level for a left turn, a straight ahead, a right turn, and a U-turn based on the confidence levels of the four driving scenarios for the obstacle. Then, based on the results of each second turning intention and a preset candidate confidence level, the device can determine the second confidence level for each direction corresponding to each second turning intention. For example, if the preset candidate confidence level is 0.5, and the result of the second turning intention for a left turn is yes, then the second confidence level for the left turn is 0.5; if the result of the second turning intention for a left turn is no, then the second confidence level for the left turn is 0.

[0108] Next, based on the first and second confidence levels for each direction, the confidence levels for the same direction are summed to determine the intention confidence level for each direction. For example, if the first confidence level for a left turn is 0.3 and the second confidence level for a left turn is 0.6, the left turn intention confidence level is 0.9. The intention confidence levels for each direction are then used as the final turning intention for the obstacle. For example, the final turning intention is (right turn: 0.7, straight ahead: 0.8, left turn: 0.5, U-turn: 0).

[0109] In addition, in one or more embodiments of the present specification, in order to more accurately determine the turning intention of the obstacle, in step S108, when determining the second confidence level corresponding to each direction, the following method may be used for determination.

[0110] Specifically, first, the gradient direction is determined according to the direction corresponding to the determined first turning intention of the obstacle, and the reference confidence is determined according to the first confidence corresponding to the first turning intention.

[0111] Next, for each direction corresponding to the second turning intention, the number of steps required to transform between that direction and the gradient direction is determined. Based on this number of steps, the gradient between the direction corresponding to the second turning intention and the gradient direction is determined. For example, for a motor vehicle, a 90-degree shift in the direction the vehicle's front is pointing left or right is considered one transformation, i.e., the number of steps required is one. For example, a motor vehicle changing from going straight to turning left requires one transformation, resulting in one step, while a motor vehicle changing from going straight to turning around requires two transformations, resulting in two steps.

[0112] Again, when determining the gradient based on the number of steps, the number of steps is proportional to the gradient. For example, the gradient between two directions determined by a transformation with a number of steps of 0 is 0, the gradient between two directions determined by a transformation with a number of steps of 1 is 1, the gradient between two directions determined by a transformation with a number of steps of 2 is 2, and the gradient between two directions determined by a transformation with a number of steps of 3 is 3.

[0113] Then, based on the different gradients, confidence reference values ​​are determined for the two directions with different gradients. Continuing with the above example, the confidence reference value for the two directions with a gradient of 0 is determined to be 1, the confidence reference value for the two directions with a gradient of 1 is determined to be 0.5, the confidence reference value for the two directions with a gradient of 2 is determined to be 0.25, and the confidence reference value for the two directions with a gradient of 3 is determined to be 0.125. Of course, other gradient reference values ​​may also be determined and set as needed, and this specification does not impose any limitations.

[0114] Finally, the second confidence level of the direction corresponding to each second turning intention is determined based on the direction corresponding to each second turning intention and the gradient of the gradient direction, the gradient reference value corresponding to each gradient, each second turning intention and the determined reference confidence level.

[0115] Continuing with the above example, when the reference confidence level is 1, the gradient direction indicates a right turn, the second intention to turn right is yes, the gradient is 0, and the gradient reference value is 1, the second confidence level for the right turn can be determined to be the product of the reference confidence level and the gradient reference value. That is, if the second confidence level for the right turn is 1 and the second intention to turn right is no, the second confidence level for the right turn can be determined to be 0. Of course, the second confidence levels corresponding to each direction can also be set using other methods. The specific method used can be determined as needed and is not limited in this specification.

[0116] In addition, in one or more embodiments of the present specification, in order to more accurately determine the turning intention of the obstacle, in step S108, when determining the gradients in any two directions, the following method may be used.

[0117] Specifically, first, based on the possible directions of the obstacle, each direction is sorted in the order of right turn, straight ahead, left turn, and U-turn. Then, based on the distance between any two directions in the sorting, the gradient between those two directions is determined. The gradient between any two directions is proportional to the distance between them in the sorting.

[0118] For example, the distance between two identical directions in the sort is determined to be 0, the gradient between any two directions with a distance of 0 is determined to be 1, the distance between any two directions separated by one direction in the sort is determined to be 1, the gradient between two directions with a distance of 1 is determined to be 2, the distance between any two directions separated by two directions in the sort is determined to be 2, and the gradient between two directions with a distance of 2 is determined to be 3.

[0119] In addition, in one or more embodiments of the present specification, in order to more accurately determine the first turning intention of the obstacle based on the HOG feature of the obstacle, in step S104, the first turning intention of the obstacle may be determined in the following manner.

[0120] Specifically, the unmanned driving device can use the determined HOG features representing the boundary of the obstacle and the trained first classifier as the input of the first classifier to obtain the output result of the first classifier as the first turning intention of the obstacle.

[0121] The first turning intention includes only one of the four driving options: right turn, straight ahead, left turn, or U-turn, and the confidence level corresponding to the driving option. For example, the first turning intention is "left turn, 0.8." Here, 0.8 is the confidence level for the left turn. Furthermore, in each subsequent step after step S106, the direction output by the first turning intention is used as the gradient direction, and the confidence level corresponding to the driving option is used as the reference confidence level. The first confidence levels for other directions are set to 0.

[0122] In addition, in one or more embodiments of the present specification, in order to more accurately determine the final turning intention of the obstacle, in step S108, when determining the final turning intention, the final turning intention may be normalized.

[0123] Specifically, according to the intention confidence of each direction, the sum of the intention confidence of each direction is determined as the total confidence. And according to the intention confidence of each direction and the total confidence, the proportion of the intention confidence of each direction in the total confidence is determined, and the proportion of the intention confidence of each direction in the total confidence is used as the final turning intention. Continuing with the above example, the right turn intention confidence is 0.7, the straight intention confidence is 0.8, the left turn intention confidence is 0.5, and the U-turn intention confidence is 0. The total confidence is 2, the right turn intention confidence accounts for 35%, the straight intention confidence accounts for 40%, the left turn intention confidence accounts for 25%, and the U-turn intention confidence accounts for 0%. The final turning intention is (right turn: 0.35, straight: 0.4, left turn: 0.25, U-turn: 0).

[0124] The above process of normalizing the final steering intention can be expressed by the formula:

[0125]

[0126]

[0127] Among them, i=1 means right turn, i=2 means going straight, i=3 means left turn, i=4 means U-turn, p i Indicates the proportion of confidence in the direction corresponding to i, e i Indicates the confidence of the direction corresponding to i, ∑e i The sum of the confidence scores for each direction can be calculated using the above formula. This allows us to normalize the final steering intent and determine the obstacle avoidance strategy.

[0128] In addition, in one or more embodiments of the present specification, in order to more accurately determine the turning intentions of each obstacle in the intersection, so as to determine the obstacle avoidance strategy for the unmanned driving device, in addition to determining the turning intentions in the four directions of right turn, straight go, left turn and U-turn, the turning intentions in other directions can also be determined.

[0129] Specifically, step S104 is described using the example of an obstacle at an intersection with only the following possible scenarios: right turn, straight ahead, left turn, and U-turn. However, in actual driving, the behavior of obstacles at intersections can be more complex, and obstacles at intersections may adopt other driving strategies. For example, turning left and merging, going straight and merging, and so on. Therefore, multiple turning intentions can be determined to more accurately determine the obstacle avoidance strategy. The specific direction of the turning intention can be determined based on actual needs and is not limited in this specification. For ease of description, this specification describes the four driving strategies where the first turning intention is limited to turning right, going straight ahead, turning left, and U-turn.

[0130] In addition, in one or more embodiments of the present specification, in order to more accurately determine the turning intention of each obstacle in the intersection, so as to determine the obstacle avoidance strategy for the unmanned driving device, the final turning intention can be determined according to the situation of the intersection.

[0131] Specifically, different intersections may have different driving directions. For example, when an unmanned vehicle reaches a three-way intersection, there are only two possible driving directions, and only the turning intentions corresponding to the two directions need to be determined. Therefore, multiple first classifiers and multiple second classifiers can be set according to different intersection types. The specific configuration of the first classifier and multiple second classifiers can be determined as needed and is not limited in this specification.

[0132] The above is a method for predicting turning intention provided by one or more embodiments of this specification. Based on the same idea, this specification also provides a corresponding device for predicting turning intention, such as Figure 4 shown.

[0133] Figure 4 A schematic diagram of a device for predicting turning intention provided in this specification, specifically including:

[0134] Acquisition module 400: used to acquire images of obstacles;

[0135] Characterization module 402: configured to extract a directional gradient histogram representing the obstacle boundary based on the image;

[0136] The first module 404 is configured to input the directional gradient histogram into a trained first classifier to obtain a first turning intention of the obstacle output by the first classifier;

[0137] The second module 406 is configured to input the directional gradient histogram of the obstacle into each trained second classifier corresponding to each turning intention, and obtain each second turning intention of the obstacle as output by each second classifier;

[0138] The determination module 408 is configured to determine a final turning intention of the obstacle according to the first turning intention and the second turning intentions, so as to determine an obstacle avoidance strategy based on the final turning intention.

[0139] Optionally, the representation module 402 performs grayscale processing on the image according to the image, and converts the image into an image of a preset size according to the grayscale processed image and a preset size.

[0140] Optionally, the characterization module 402 divides the image into a plurality of regions. For each region, based on preset gradient directions, a feature vector corresponding to each gradient direction in the region is determined, and based on the feature vectors corresponding to each gradient direction in each region, a directional gradient histogram representing the obstacle boundary is determined.

[0141] Optionally, the first module 404 uses the directional gradient histogram and the trained first classifier as the input of the first classifier to obtain the output result of the first classifier as the first turning intention of the obstacle.

[0142] Optionally, the second module 406, based on the directional gradient histogram and the trained right-turn second classifier, straight-ahead second classifier, left-turn second classifier, and U-turn second classifier, takes the directional gradient histogram as input, and determines the output of the right-turn second classifier as the right-turn second turning intention of the obstacle, the output of the straight-ahead second classifier as the straight-ahead second turning intention of the obstacle, the output of the left-turn second classifier as the left-turn second turning intention of the obstacle, and the output of the U-turn second classifier as the U-turn second turning intention of the obstacle.

[0143] Optionally, the determination module 408 determines a first confidence level of the direction corresponding to the first steering intention based on the first steering intention, determines a second confidence level of each direction corresponding to each second steering intention based on each second steering intention, and determines the final steering intention of the obstacle based on the first confidence level of the direction corresponding to the first steering intention and the second confidence level of each direction corresponding to each second steering intention.

[0144] Optionally, the determination module 408 determines the gradient direction based on the direction corresponding to the first steering intention, determines the reference confidence based on the first confidence corresponding to the first steering intention, determines the number of steps for transforming between the direction and the gradient direction for each direction corresponding to the second steering intention, determines the gradient of the direction corresponding to the second steering intention and the gradient direction based on the number of steps, and determines the second confidence of the direction corresponding to each second steering intention based on each second steering intention, the gradient of the direction corresponding to each second steering intention and the gradient direction, and the reference confidence.

[0145] This specification also provides a computer-readable storage medium, which stores a computer program that can be used to execute the above Figure 1 A method for predicting steering intention is provided.

[0146] This manual also provides Figure 5The schematic structural diagram of the unmanned driving equipment is shown in FIG. Figure 5 As mentioned above, at the hardware level, the unmanned driving device includes a processor, an internal bus, a network interface, a memory and a non-volatile memory, and may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0147] In the 1990s, technological improvements could be clearly distinguished as either hardware improvements (for example, improvements to circuit structures like diodes, transistors, and switches) or software improvements (improvements to process flows). However, with the advancement of technology, many process flow improvements today can now be considered direct improvements to hardware circuit structures. Designers almost always create the corresponding hardware circuit structure by programming the improved process flow into the hardware circuit. Therefore, it cannot be said that a process flow improvement cannot be implemented using hardware modules. For example, a programmable logic device (PLD), such as a field programmable gate array (FPGA), is an integrated circuit whose logical function is determined by user programming. Designers can "integrate" a digital system on a PLD through their own programming, without having to hire a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly done using "logic compiler" software. This is similar to the software compiler used when developing programs. Before compilation, the original code must also be written in a specific programming language, called a hardware description language (HDL). There is not just one HDL, but many, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art will also understand that by simply programming the method flow in one of these hardware description languages ​​and then programming it into an integrated circuit, a hardware circuit that implements the logic method flow can be easily obtained.

[0148] The controller can be implemented in any suitable manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that in addition to implementing the controller in a purely computer-readable program code format, the controller can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules that implement the method and structures within the hardware component.

[0149] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0150] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0151] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0152] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0153] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0154] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0155] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0156] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0157] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0158] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0159] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Thus, this specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0160] This specification may be described in the general context of computer-executable instructions, such as program modules, executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media, including storage devices.

[0161] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0162] The foregoing is merely an embodiment of the present invention and is not intended to limit the present invention. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of the claims of this application.

Claims

1. A method for predicting turning intention, characterized in that: include: Obtain an image of the obstacle; Extracting a directional gradient histogram representing the boundary of the obstacle according to the image; Inputting the directional gradient histogram into a trained first classifier to obtain a first turning intention of the obstacle output by the first classifier; Inputting the directional gradient histogram of the obstacle into each second classifier corresponding to each trained turning intention, respectively, to obtain each second turning intention of the obstacle as output by each second classifier; determining a final turning intention of the obstacle according to the first turning intention and the second turning intentions, and determining an obstacle avoidance strategy based on the final turning intention; The determining, based on the first turning intention and the second turning intentions, a final turning intention of the obstacle specifically includes: determining, according to the first turning intention, a first confidence level of a direction corresponding to the first turning intention; determining, according to each second turning intention, a second confidence level for each direction corresponding to each second turning intention; Determining a final turning intention of the obstacle based on a first confidence level of the direction corresponding to the first turning intention and a second confidence level of each direction corresponding to each second turning intention; the method further comprising: determining a gradient direction according to the direction corresponding to the first turning intention, and determining a reference confidence level according to a first confidence level corresponding to the first turning intention; For each direction corresponding to the second turning intention, determining a number of steps for transforming the direction into the gradient direction, and determining a gradient between the direction corresponding to the second turning intention and the gradient direction according to the number of steps; A second confidence level for the direction corresponding to each second turning intention is determined according to each second turning intention, the direction corresponding to each second turning intention, the gradient of the gradient direction, and the reference confidence level.

2. The method according to claim 1, wherein Before extracting the directional gradient histogram representing the obstacle boundary based on the image, the method further includes: According to the image, performing grayscale processing on the image; According to the grayscale-processed image and the preset size, the image is converted into an image with a preset size.

3. The method according to claim 1, wherein Extracting a directional gradient histogram representing the obstacle boundary based on the image specifically includes: According to the image, dividing the image into a plurality of regions; For each region, according to the preset gradient directions, determine the eigenvectors corresponding to the gradient directions in the region; A directional gradient histogram representing the boundary of the obstacle is determined according to the feature vector corresponding to each gradient direction in each area.

4. The method according to claim 1, wherein Inputting the directional gradient histogram into a trained first classifier to obtain a first turning intention of the obstacle output by the first classifier specifically includes: According to the directional gradient histogram and the trained first classifier, the directional gradient histogram is used as the input of the first classifier, and the output result of the first classifier is obtained as the first turning intention of the obstacle.

5. The method according to claim 1, wherein Inputting the directional gradient histogram of the obstacle into each trained second classifier corresponding to each turning intention, and obtaining each second turning intention of the obstacle output by each second classifier, specifically includes: According to the directional gradient histogram, as well as the trained right-turn second classifier, straight-ahead second classifier, left-turn second classifier, and U-turn second classifier, the directional gradient histogram is used as input to determine the output of the right-turn second classifier as the right-turn second turning intention of the obstacle, the output of the straight-ahead second classifier as the straight-ahead second turning intention of the obstacle, the output of the left-turn second classifier as the left-turn second turning intention of the obstacle, and the output of the U-turn second classifier as the U-turn second turning intention of the obstacle.

6. A device for predicting turning intention, characterized in that: include: Acquisition module: used to acquire images of obstacles; A characterization module is configured to extract a directional gradient histogram representing the boundary of the obstacle based on the image; The first module is configured to input the directional gradient histogram into a trained first classifier to obtain a first turning intention of the obstacle output by the first classifier; The second module is configured to input the directional gradient histogram of the obstacle into each second classifier corresponding to each trained turning intention, and obtain each second turning intention of the obstacle as output by each second classifier; a determination module configured to determine a final turning intention of the obstacle according to the first turning intention and each of the second turning intentions, so as to determine an obstacle avoidance strategy based on the final turning intention; The determining, based on the first turning intention and the second turning intentions, a final turning intention of the obstacle specifically includes: determining, according to the first turning intention, a first confidence level of a direction corresponding to the first turning intention; determining, according to each second turning intention, a second confidence level for each direction corresponding to each second turning intention; determining a final turning intention of the obstacle based on a first confidence level of the direction corresponding to the first turning intention and a second confidence level of each direction corresponding to each second turning intention; Also includes: determining a gradient direction according to the direction corresponding to the first turning intention, and determining a reference confidence level according to a first confidence level corresponding to the first turning intention; For each direction corresponding to the second turning intention, determining a number of steps for transforming the direction into the gradient direction, and determining a gradient between the direction corresponding to the second turning intention and the gradient direction according to the number of steps; A second confidence level for the direction corresponding to each second turning intention is determined according to each second turning intention, the direction corresponding to each second turning intention, the gradient of the gradient direction, and the reference confidence level.

7. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

8. An unmanned driving device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 5 is implemented.

Images