Method, device, equipment and storage medium for identifying the distance between a vehicle and an obstacle

Abstract

The present application relates to a method, device, equipment and storage medium for identifying the distance between a vehicle and an obstacle. The method includes: obtaining point cloud data and / or image data of a current frame of a current obstacle; obtaining the observed position information of at least one side of the current frame of the current obstacle, the uncertainty of each side and the predicted information of the center position based on the point cloud data and / or image data; obtaining predicted size information based on the observed position information of each side, the uncertainty of each side and the predicted information of the center position; determining the position information of a target rectangle based on the predicted size information and the predicted information of the center position; determining the distance between the current obstacle and the current vehicle in the current frame based on the position information of the target rectangle. The present application obtains the predicted size information of the current frame of the current obstacle based on the predicted information of the center position, so that the distance between the vehicle and the current obstacle obtained based on the predicted size information is more accurate.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, equipment and storage medium for identifying the distance between a vehicle and an obstacle. Background Art

[0002] In the field of autonomous driving, it is often necessary to calculate the distance between the autonomous driving vehicle and obstacles.

[0003] The traditional approach is to use a deep learning model to directly estimate the distance from the geometric center of the obstacle vehicle to the target vehicle. The method used to calculate the collision is to directly fit a rectangle of predefined length and width to the center according to the identified vehicle type, such as a car, a truck, or a large truck, and then calculate the collision.

[0004] The rectangle is predefined, for example, it is simply divided into large vehicles (>13m), medium vehicles (13-6m) and small vehicles (<6m), etc., rather than the actual vehicle size, which may have a large error in some cases, such as in front of the vehicle. At the same time, the geometric center of the obstacle vehicle we detect is not necessarily the real center. At this time, if the predefined rectangle is still applied, it will further lead to inaccurate estimated distances, especially when following a vehicle at a close distance, which may cause rear-end collisions and other problems. Summary of the invention

[0005] Based on this, it is necessary to provide a method, device, equipment and storage medium for identifying the distance between a vehicle and an obstacle in order to solve the above technical problems.

[0006] A method for identifying the distance between a vehicle and an obstacle, the method comprising:

[0007] Obtaining point cloud data and / or image data of a current frame of a current obstacle;

[0008] Obtaining, according to the point cloud data and / or the image data, the observed position information of at least one side of the current obstacle in the current frame, the uncertainty of each side, and the predicted information of the center position;

[0009] According to the observed position information of each side, the uncertainty of each side and the predicted information of the center position, the predicted size information of the current frame of the current obstacle is obtained;

[0010] Determine the position information of the target rectangle according to the predicted size information and the predicted center position information;

[0011] According to the position information of the target rectangle, the distance between the current obstacle and the current vehicle in the current frame is determined.

[0012] In one embodiment, the above-mentioned obtaining the predicted size information of the current frame of the current obstacle according to the observed position information of each side, the uncertainty of each side and the predicted information of the center position includes:

[0013] Input the position information of each edge, the uncertainty of each edge and the predicted information of the center position into the target filter;

[0014] According to the prediction information of the center position and the current internal state data of the target filter, the converted position information of each side of the current frame is obtained, and the current internal state data is obtained according to the internal state data of the target filter of the previous frame and the position information of each side of the current frame;

[0015] Determine the distance between the converted position information of each edge and the corresponding observed position information according to the converted position information of each edge of the current frame and the observed position information of each edge;

[0016] According to the distance between the converted position information of each edge and the corresponding observed position information and the uncertainty of each edge, the predicted size information of the obstacle corresponding to the current frame is obtained.

[0017] In one embodiment, obtaining the observed position information of at least one side of the current obstacle in the current frame, the uncertainty of each side, and the predicted information of the center position according to the point cloud data and / or the image data includes:

[0018] The point cloud data is segmented using a preset first convolutional neural network, and the segmented point cloud data is regressed using a preset second convolutional neural network to obtain prediction information of the center position and observation position information of each side;

[0019] Determine the density of the point cloud within a preset range around each edge according to the observation position information of each edge and the point cloud data;

[0020] The uncertainty of each edge is determined based on the density of the point cloud within a preset range around each edge.

[0021] In one embodiment, the predicted size information of the obstacle corresponding to the current frame is obtained according to the distance between the converted position information of each edge and the corresponding observed position information and the uncertainty of each edge, including:

[0022] Construct a covariance matrix based on the uncertainty on each side;

[0023] According to the covariance matrix, the distance between the converted position information of each edge and the corresponding observed position information, and the unscented Kalman filter algorithm, the predicted size information of the obstacle corresponding to the current frame is determined.

[0024] In one embodiment, the method further comprises:

[0025] Determine the orientation information of the current frame of the current obstacle according to the point cloud data and / or the image data;

[0026] The above-mentioned conversion position information of each side of the current frame is obtained based on the prediction information of the center position and the current internal state data of the target filter, including:

[0027] According to the prediction information of the center position, the current internal state data of the target filter and the orientation information, the converted position information of each side of the current frame is obtained.

[0028] In one embodiment, the above-mentioned determining the position information of the target rectangle according to the predicted size information and the predicted center position information includes:

[0029] The position information of the target rectangle is determined based on the predicted size information, the predicted center position information, and the orientation information.

[0030] In one embodiment, the step of obtaining point cloud data and / or image data of a current frame of a current obstacle includes:

[0031] The point cloud data and / or image data are collected by multiple sensors.

[0032] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0033] Acquire point cloud data and / or image data of a current frame of a current obstacle, the point cloud data and / or image data comprising point cloud data and / or image data;

[0034] Obtaining, according to the point cloud data and / or the image data, the observed position information of at least one side of the current obstacle in the current frame, the uncertainty of each side, and the predicted information of the center position;

[0035] According to the observed position information of each side, the uncertainty of each side and the predicted information of the center position, the predicted size information of the current frame of the current obstacle is obtained;

[0036] Determine the position information of the target rectangle according to the predicted size information and the predicted center position information;

[0037] According to the position information of the target rectangle, the distance between the current obstacle and the current vehicle in the current frame is determined.

[0038] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0039] Acquire point cloud data and / or image data of a current frame of a current obstacle, the point cloud data and / or image data comprising point cloud data and / or image data;

[0040] Obtaining, according to the point cloud data and / or the image data, the observed position information of at least one side of the current obstacle in the current frame, the uncertainty of each side, and the predicted information of the center position;

[0041] According to the observed position information of each side, the uncertainty of each side and the predicted information of the center position, the predicted size information of the current frame of the current obstacle is obtained;

[0042] Determine the position information of the target rectangle according to the predicted size information and the predicted center position information;

[0043] According to the position information of the target rectangle, the distance between the current obstacle and the current vehicle in the current frame is determined.

[0044] The above-mentioned method, device, equipment and storage medium for identifying the distance between a vehicle and an obstacle obtains point cloud data and / or image data of a current frame of a current obstacle; obtains observed position information of at least one side of the current frame of the current obstacle, uncertainty of each side and predicted information of the center position based on the point cloud data and / or image data; obtains predicted size information of the current frame of the current obstacle based on the observed position information of each side, uncertainty of each side and predicted information of the center position; determines the position information of a target rectangle based on the predicted size information and predicted information of the center position; and determines the distance between the current obstacle and the current vehicle in the current frame based on the position information of the target rectangle. The present application can obtain the predicted size information of the current frame of the current obstacle through the observed position information of each side, the uncertainty of each side and the predicted information of the center position. The predicted size information is calculated based on the predicted information of the center position. Therefore, the predicted size information is adapted to the predicted information of the center position. This adaptation can make the subsequently determined target rectangle closer to its actual position, making the calculated distance more accurate. Compared with the traditional technology, the center position of the obstacle is estimated, and the size information adopts preset specifications. The preset size specifications are set on the estimated center position to calculate the distance between the vehicle and the obstacle, which can easily lead to the problem of inaccurate calculated distance. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A schematic flow chart of a method for identifying the distance between a vehicle and an obstacle in one embodiment;

[0046] Figure 2It is a flowchart of a step of obtaining predicted size information of a current frame of a current obstacle according to position information of at least one edge, uncertainty of each of the above edges and predicted information of a center position in an embodiment;

[0047] Figure 3 A flowchart of the steps of obtaining the observed position information of at least one side of the current obstacle in the current frame, the uncertainty of each side, and the predicted information of the center position according to the point cloud data and / or the image data in one embodiment;

[0048] Figure 4 is a structural block diagram of a device for identifying the distance between a vehicle and an obstacle in one embodiment;

[0049] Figure 5 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0051] In one embodiment, Figure 1 As shown, a method for identifying the distance between a vehicle and an obstacle is provided, and the method is applied to a server as an example for explanation, and includes the following steps:

[0052] S11. Obtain point cloud data and / or image data of a current frame of a current obstacle.

[0053] In the present application, the above-mentioned current obstacle may be a vehicle or other obstacle. The above-mentioned point cloud data may be data collected by a laser radar or a millimeter wave radar. The above-mentioned image data may be data collected by a camera. The point cloud data and / or image data in the present application may also be data collected by other sensors, which are not specifically limited here.

[0054] S12. Obtaining, based on the point cloud data and / or the image data, the observed position information of at least one side of the current frame of the current obstacle, the uncertainty of each side, and the predicted information of the center position.

[0055] In the present application, the observed position information of the edge is the position information calculated based on the point cloud data and / or image data. The uncertainty can be the degree of difference between the observed position information of the edge and the actual position information. The predicted information of the center position can be the position information obtained by predicting the geometric center point of the current obstacle.

[0056] In the present application, the collected point cloud data and / or image data can be processed, for example, the point cloud data and / or image data can be processed using a neural network. Specifically, for example, the observed position information of at least one edge, the uncertainty of each edge, and the predicted information of the center position can be obtained by processing the point cloud data and / or image data using processing techniques such as point cloud segmentation or image segmentation.

[0057] S13. Obtain the predicted size information of the current frame of the current obstacle based on the observed position information of each side, the uncertainty of each side and the predicted information of the center position.

[0058] In the present application, the above-mentioned predicted size information may be the predicted length and predicted width of the current frame of the current obstacle.

[0059] In the present application, the observed position information of each side, the uncertainty of each side and the predicted information of the center position can be input into a preset filter, and the observed position information of each side, the uncertainty of each side and the predicted information of the center position can be processed by the filter to obtain the predicted size information of the current frame of the current obstacle.

[0060] The present application can also input the observed position information of each side, the uncertainty of each side, and the predicted information of the center position into a pre-constructed neural network, and process the observed position information of each side, the uncertainty of each side, and the predicted information of the center position through the neural network to obtain the predicted size information of the current frame of the current obstacle.

[0061] S14. Determine the position information of the target rectangle according to the predicted size information and the predicted center position information.

[0062] In the present application, the target rectangle is a rectangle obtained by predicting the shape of the current obstacle. The position information of the target rectangle may include the position information of each side of the target rectangle, or may also include the position information of the center point of the target rectangle.

[0063] The present application can construct a coordinate system and determine the position information of the target rectangle in the coordinate system based on the predicted size information and the predicted center position information.

[0064] S15. Determine the distance between the current obstacle and the current vehicle in the current frame according to the position information of the target rectangle.

[0065] In the present application, the position information of the target rectangle and the position information of the current vehicle can be determined in advance in the coordinate system, and the position information of the point closest to the current vehicle on each side of the target rectangle can be determined. Based on the distance from the position information of the closest point to the current vehicle, the distance between the current obstacle and the current vehicle in the current frame can be obtained.

[0066] In one embodiment, if Figure 2 As shown, the above-mentioned obtaining the predicted size information of the current frame of the current obstacle according to the position information of at least one edge, the uncertainty of each of the above-mentioned edges and the predicted information of the center position may include:

[0067] S21, inputting the position information of each side, the uncertainty of each side and the prediction information of the center position into the target filter;

[0068] S22, obtaining the converted position information of each side of the current frame according to the prediction information of the center position and the current internal state data of the target filter, wherein the current internal state data is obtained according to the internal state data of the target filter of the previous frame and the position information of each side of the current frame;

[0069] S23, determining the distance between the converted position information of each edge and the corresponding observed position information according to the converted position information of each edge of the current frame and the observed position information of each edge;

[0070] S24. Obtain the predicted size information of the obstacle corresponding to the current frame according to the distance between the converted position information of each edge and the corresponding observed position information and the uncertainty of each edge.

[0071] In the present application, the above-mentioned target filter may include an unscented Kalman filter, an extended Kalman filter, a linear Kalman filter, and the like.

[0072] In this application, the internal state data of the target filter needs to be initialized in advance. The initialized target filter is then used to process the observed position information of each edge corresponding to each frame, the uncertainty of each edge, and the predicted information of the center position. When processing the data of the tth frame, the internal state data of the target filter corresponding to the tth frame can be determined by the following method:

[0073] Obtain the internal state data of the target filter corresponding to the t-1th frame;

[0074] Get the observed position information and uncertainty of each edge corresponding to the tth frame;

[0075] According to the internal state data of the target filter corresponding to the t-1th frame and the observed position information of each edge corresponding to the tth frame and the uncertainty of each edge, the internal state data of the target filter corresponding to the tth frame is obtained.

[0076] The above-mentioned prediction information of the center position may be the predicted coordinate value of the center position. The above-mentioned obtaining the converted position information of each side of the current frame according to the prediction information of the center position and the current internal state data of the target filter may include:

[0077] According to the current internal state data of the target filter and the predicted coordinate value of the center position, the coordinate value of the midpoint of each edge is obtained;

[0078] According to the coordinate value of the midpoint of each edge and the direction vector of each edge, the converted position information of each edge is obtained.

[0079] The direction vectors of the aforementioned edges may be determined based on the orientation information of the vehicle body in the current frame, or may be determined based on the orientation information of the current lane line.

[0080] The above-mentioned determining the distance between the converted position information of each edge and the corresponding observed position information according to the converted position information of each edge of the current frame and the observed position information of each edge may include:

[0081] Determine a first rectangle according to the converted position information of each side;

[0082] Determine a second rectangle based on the observed position information of each side;

[0083] The distances from points on each side of the first rectangle to the corresponding sides of the second rectangle are calculated to obtain the distances between the converted position information of each side and the corresponding observed position information.

[0084] The predicted size information of the obstacle corresponding to the current frame obtained based on the distance between the converted position information of each edge and the corresponding observed position information and the uncertainty of each edge can be determined by a preset algorithm.

[0085] In the present application, when the program starts, the internal state data of the target filter needs to be initialized, and the predicted size information of the first frame is subsequently calculated based on the value initialized by the internal state data, and the internal state data of the target filter corresponding to the second frame is obtained based on the internal state data of the target filter of the first frame, and the predicted size information of the second frame is determined based on the internal state data of the second frame, and so on. Therefore, the internal state data of the target filter is constantly updated, and its internal state data at any time is obtained by integrating the predicted size information of all previous frames, and includes the information of all previous frames.

[0086] Through this implementation, the present application can convert the observation values ​​of each side into observations of the length and width of each side through a target filter, and the predicted size information of the obstacle is calculated based on the predicted information of the center position. The predicted size information obtained in this way and the target rectangle determined by the predicted information of the center position are closer to its actual position, so that the final distance between the current vehicle and the obstacle is also more accurate.

[0087] In one embodiment, if Figure 3As shown, the above-mentioned obtaining the observed position information of at least one side of the current frame of the current obstacle, the uncertainty of each side and the predicted information of the center position according to the point cloud data and / or image data may include:

[0088] S31, segmenting the point cloud data using a preset first convolutional neural network, and performing circumscribed rectangle regression on the segmented point cloud data using a preset second convolutional neural network to obtain prediction information of the center position and observation position information of each side;

[0089] S32, determining the density of the point cloud within a preset range around each edge according to the observation position information of each edge and the point cloud data;

[0090] S33. Determine the uncertainty of each edge according to the density of the point cloud within a preset range around each edge.

[0091] In the present application, the first convolutional neural network and the second convolutional neural network can be CNN (Convolutional Neural Networks). In the present application, the center position information of the circumscribed rectangle obtained by performing circumscribed rectangle regression using the second convolutional neural network is the prediction information of the center position of the current obstacle, and the position information of the four sides of the circumscribed rectangle is the position information of the four sides of the current obstacle.

[0092] In this application, the density of the point cloud within the preset range on both sides of each side can be determined based on the observed position information of each side and the point cloud data, and the uncertainty of each side can be determined based on the obtained densities. The greater the difference in density of the point cloud within the preset range on both sides, the lower the uncertainty, and vice versa.

[0093] Through this implementation, the present application can predict the center position of the current obstacle and determine the uncertainty of each side, so as to subsequently determine the density of the point cloud within a preset range around each side based on the observed position information of each side and the point cloud data, and obtain the predicted size information of the current frame of the current obstacle, so that the obtained predicted size information is adapted to the center position, and even if the center position is not accurately estimated, the accurate distance between the current obstacle and the current vehicle can be obtained.

[0094] In one embodiment, obtaining the predicted size information of the obstacle corresponding to the current frame according to the distance between the converted position information of each edge and the corresponding observed position information and the uncertainty of each edge may include:

[0095] Construct a covariance matrix based on the uncertainty on each side;

[0096] According to the covariance matrix, the distance between the converted position information of each edge and the corresponding observed position information, and the unscented Kalman filter algorithm, the predicted size information of the obstacle corresponding to the current frame is determined.

[0097] In this application, assuming that the observed position information of four edges is obtained, the above-mentioned covariance matrix constructed according to the uncertainty of each edge may include:

[0098] According to the four uncertainties of the strip edges, a 4x4 covariance matrix is ​​constructed, and the obtained covariance matrix can be in the following form;

[0099]

[0100] Among them, the present application can also obtain other numbers of edges, not limited to four. For example, in one embodiment, it is also possible to obtain one edge, and then construct a 1x1 covariance matrix. Similarly, it is also possible to obtain two edges or three edges. The above-mentioned unscented Kalman filter algorithm is an algorithm that combines data changes (UT transformation) with Kalman filtering. Through the UT transformation, a state point is converted into multiple state points, which are passed to the subsequent observations according to certain weights, and then according to the error between the subsequent observations and the true value, feedback is fed forward, and corrections are continuously made to finally obtain the ideal state value.

[0101] The above-mentioned unscented Kalman filter algorithm includes a conversion function, which is used to convert the current internal state data of the target filter and the predicted information of the center position into the converted position information of the four sides, and then calculate the distance from the observed position information of each side. The unscented Kalman filter algorithm uses this distance and the uncertainty of each side to optimize and update the internal state data of the target filter, and uses the updated internal state data of the target filter as the predicted size information of the obstacle corresponding to the current frame.

[0102] The present application uses an unscented Kalman filter algorithm to determine the predicted size information of the obstacle corresponding to the current frame, thereby converting the observation of the position information of each side into the observation of the length and width of each side, and determining the target rectangle based on the obtained length and width of each side and the predicted information of the center position, thereby solving the problem of inaccurate calculation results caused by determining the target rectangle through the predicted information of the center position and the preset length and width specifications in the traditional technology.

[0103] In one embodiment, the above method may further include:

[0104] Determine the orientation information of the current frame of the current obstacle according to the point cloud data and / or the image data;

[0105] The above-mentioned obtaining the converted position information of each side of the current frame according to the prediction information of the center position and the current internal state data of the target filter may include:

[0106] According to the prediction information of the center position, the current internal state data of the target filter and the orientation information, the converted position information of each side of the current frame is obtained.

[0107] In the present application, when the above-mentioned obstacle is a vehicle, the above-mentioned orientation information may be the vehicle body orientation information of the current obstacle, and the vehicle body orientation information may be the direction in which the vehicle body is pointing.

[0108] The above-mentioned obtaining the converted position information of each side of the current frame according to the prediction information of the center position, the current internal state data of the target filter and the orientation information may include:

[0109] According to the length and width of the obstacle in the previous frame and the predicted coordinate values ​​of the center position, the coordinate values ​​of the midpoints of each edge are obtained;

[0110] Determine the direction vector of each edge according to the vehicle body orientation information;

[0111] According to the coordinate value of the midpoint of each edge and the direction vector of each edge, the converted position information of each edge is obtained.

[0112] In one embodiment, the above-mentioned determining the position information of the target rectangle according to the predicted size information and the predicted center position information may include:

[0113] The position information of the target rectangle is determined based on the predicted size information, the predicted center position information, and the orientation information.

[0114] In the present application, the above-mentioned predicted size information may be the length and width of the current obstacle. The above-mentioned direction vectors of each side of the obstacle may be determined based on the orientation information, and the position information of the target rectangle may be determined based on the direction vectors of each side and the length and width of the current obstacle and the predicted information of the center position.

[0115] In one embodiment, the above-mentioned obtaining of the point cloud data and / or image data of the current frame of the current obstacle may include:

[0116] The point cloud data and / or image data are collected by multiple sensors.

[0117] In this application, point cloud data and / or image data from multiple sensors of different sources may be accepted, and the sensors may include cameras, laser radars, millimeter wave radars, etc. Multiple sensors observe the current obstacle at different angles or dimensions to obtain point cloud data and / or image data.

[0118] Specifically, the present application collects point cloud data of the current obstacle through a laser radar. The point cloud data will pass through a 3D segmentation neural network to segment the point cloud belonging to the current obstacle, and then connect a small convolutional neural network to regress the bounding rectangle of the current obstacle. The coordinate values ​​of the four sides of the bounding rectangle are obtained. The uncertainty of each side can be output by the convolutional neural network itself, or it can be converted based on features such as the point cloud data of the current obstacle. Among them, the more and denser the points, the more accurate the bounding rectangle obtained, the lower the uncertainty, and vice versa.

[0119] This application can also collect point cloud data and / or image data through millimeter wave radar, similar to laser radar, which first obtains point cloud data and then performs segmentation through the model. The difference is that the point cloud of millimeter wave radar is very sparse, and it is usually impossible to directly regress the circumscribed rectangle of the object. Therefore, we will only regress those reliable edges (fast speed, no obstruction) and give a fixed uncertainty. The uncertainty of each edge is determined by the speed.

[0120] This application can also collect point cloud data and / or image data through cameras. Because there are multiple cameras with different viewing angles and focal lengths, each one sees something different. First, we will regress the bounding rectangle of the current obstacle through the CNN convolutional data network within each camera's own field of view. Note that when the current obstacle appears in the field of view of more than one camera, multiple bounding rectangles will be observed. At this time, we need to select the 4 edges that are finally provided for tracking. They may come from different rectangles observed by different cameras. Generally speaking, data observed by cameras with small uncertainty is selected. As for the uncertainty of the camera, it can be set manually. For example, a camera with a 120-degree viewing angle will definitely not see distant objects as clearly as a camera with a 60-degree viewing angle, and the uncertainty is correspondingly greater.

[0121] In one embodiment, Figure 4 As shown, a device for identifying the distance between a vehicle and an obstacle is provided, comprising: an acquisition module 11, a first processing module 12, a second processing module 13, a first determination module 14 and a second determination module 15, wherein:

[0122] An acquisition module 11 is used to acquire point cloud data and / or image data of a current frame of a current obstacle;

[0123] The first processing module 12 is used to obtain the observed position information of at least one side of the current obstacle in the current frame, the uncertainty of each side and the predicted information of the center position according to the point cloud data and / or the image data;

[0124] The second processing module 13 is used to obtain the predicted size information of the current frame of the current obstacle according to the observed position information of each side, the uncertainty of each side and the predicted information of the center position;

[0125] A first determination module 14 is used to determine the position information of the target rectangle according to the predicted size information and the predicted center position information;

[0126] The second determination module 15 is used to determine the distance between the current obstacle and the current vehicle in the current frame according to the position information of the target rectangle.

[0127] In one embodiment, the second processing module 13 can input the position information of each edge, the uncertainty of each edge and the predicted information of the center position into the target filter, and obtain the converted position information of each edge of the current frame according to the predicted information of the center position and the current internal state data of the target filter, the current internal state data is obtained according to the internal state data of the target filter of the previous frame and the position information of each edge of the current frame, and the distance between the converted position information of each edge and the corresponding observed position information is determined according to the converted position information of each edge of the current frame and the observed position information of each edge, and the predicted size information of the obstacle corresponding to the current frame is obtained according to the distance between the converted position information of each edge and the corresponding observed position information and the uncertainty of each edge.

[0128] In one embodiment, the first processing module 12 can use a preset first convolutional neural network to segment the point cloud data, and use a preset second convolutional neural network to perform circumscribed rectangle regression on the segmented point cloud data to obtain prediction information of the center position and observation position information of each edge, determine the density of the point cloud within a preset range around each edge based on the observation position information and the point cloud data of each edge, and determine the uncertainty of each edge based on the density of the point cloud within a preset range around each edge.

[0129] In one embodiment, the second processing module 13 can also construct a covariance matrix based on the uncertainty of each edge, and determine the predicted size information of the obstacle corresponding to the current frame based on the covariance matrix, the distance between the converted position information of each edge and the corresponding observed position information and the unscented Kalman filter algorithm.

[0130] In one embodiment, the above-mentioned device for identifying the distance between the vehicle and the obstacle also includes a third determination module (not shown in the figure), which can determine the orientation information of the current frame of the current obstacle based on the point cloud data and / or image data. The above-mentioned second processing module 13 can also obtain the converted position information of each edge of the current frame based on the prediction information of the center position, the current internal state data of the target filter and the orientation information.

[0131] In one embodiment, the first determination module 14 may determine the position information of the target rectangle according to the predicted size information, the predicted center position information and the orientation information.

[0132] In one embodiment, the acquisition module 11 may collect point cloud data and / or image data through multiple sensors.

[0133] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as the operation data of smart home devices. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a resource allocation method for compiling a virtual machine is implemented.

[0134] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program: acquiring point cloud data and / or image data of a current frame of a current obstacle; obtaining observed position information of at least one edge of the current frame of the current obstacle, uncertainty of each edge, and predicted information of a center position based on the point cloud data and / or image data; obtaining predicted size information of the current frame of the current obstacle based on the observed position information of each edge, uncertainty of each edge, and predicted information of the center position; determining position information of a target rectangle based on the predicted size information and predicted information of the center position; and determining a distance between the current obstacle and the current vehicle in the current frame based on the position information of the target rectangle.

[0135] In one embodiment, when the processor executes the computer program to implement the above-mentioned step of obtaining the predicted size information of the current frame of the current obstacle according to the observed position information of each side, the uncertainty of each side and the predicted information of the center position, the following steps are specifically implemented:

[0136] Input the position information of each edge, the uncertainty of each edge and the predicted information of the center position into the target filter;

[0137] According to the prediction information of the center position and the current internal state data of the target filter, the converted position information of each side of the current frame is obtained, and the current internal state data is obtained according to the internal state data of the target filter of the previous frame and the position information of each side of the current frame;

[0138] Determine the distance between the converted position information of each edge and the corresponding observed position information according to the converted position information of each edge of the current frame and the observed position information of each edge;

[0139] According to the distance between the converted position information of each edge and the corresponding observed position information and the uncertainty of each edge, the predicted size information of the obstacle corresponding to the current frame is obtained.

[0140] In one embodiment, when the processor executes the computer program to implement the above-mentioned step of obtaining the observed position information of at least one edge of the current frame of the current obstacle, the uncertainty of each edge, and the predicted information of the center position according to the point cloud data and / or image data, the following steps are specifically implemented:

[0141] The point cloud data is segmented using a preset first convolutional neural network, and the segmented point cloud data is regressed using a preset second convolutional neural network to obtain prediction information of the center position and observation position information of each side;

[0142] Determine the density of the point cloud within a preset range around each edge according to the observation position information of each edge and the point cloud data;

[0143] The uncertainty of each edge is determined based on the density of the point cloud within a preset range around each edge.

[0144] In one embodiment, when the processor executes the computer program to implement the above-mentioned step of obtaining the predicted size information of the obstacle corresponding to the current frame according to the distance between the converted position information of each edge and the corresponding observed position information and the uncertainty of each edge, the following steps are specifically implemented:

[0145] Construct a covariance matrix based on the uncertainty on each side;

[0146] According to the covariance matrix, the distance between the converted position information of each edge and the corresponding observed position information, and the unscented Kalman filter algorithm, the predicted size information of the obstacle corresponding to the current frame is determined.

[0147] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0148] Determine the orientation information of the current frame of the current obstacle according to the point cloud data and / or the image data;

[0149] When the processor executes the computer program to implement the above-mentioned step of obtaining the converted position information of each side of the current frame according to the prediction information of the center position and the current internal state data of the target filter, the following steps are specifically implemented:

[0150] According to the prediction information of the center position, the current internal state data of the target filter and the orientation information, the converted position information of each side of the current frame is obtained.

[0151] In one embodiment, when the processor executes the computer program to implement the above step of determining the position information of the target rectangle according to the predicted size information and the predicted information of the center position, the following steps are specifically implemented:

[0152] The position information of the target rectangle is determined based on the predicted size information, the predicted center position information, and the orientation information.

[0153] In one embodiment, when the processor executes the computer program to implement the above-mentioned step of obtaining the point cloud data and / or image data of the current frame of the current obstacle, the following steps are specifically implemented:

[0154] The point cloud data and / or image data are collected by multiple sensors.

[0155] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: acquiring point cloud data and / or image data of a current frame of a current obstacle; obtaining observed position information of at least one edge of the current frame of the current obstacle, uncertainty of each edge, and predicted information of a center position based on the point cloud data and / or image data; obtaining predicted size information of the current frame of the current obstacle based on the observed position information of each edge, uncertainty of each edge, and predicted information of the center position; determining position information of a target rectangle based on the predicted size information and the predicted information of the center position; and determining a distance between the current obstacle and the current vehicle in the current frame based on the position information of the target rectangle.

[0156] In one embodiment, when the computer program is executed by a processor to implement the above-mentioned step of obtaining the predicted size information of the current frame of the current obstacle according to the observed position information of each side, the uncertainty of each side and the predicted information of the center position, the following steps are specifically implemented:

[0157] Input the position information of each edge, the uncertainty of each edge and the predicted information of the center position into the target filter;

[0158] According to the prediction information of the center position and the current internal state data of the target filter, the converted position information of each side of the current frame is obtained, and the current internal state data is obtained according to the internal state data of the target filter of the previous frame and the position information of each side of the current frame;

[0159] Determine the distance between the converted position information of each edge and the corresponding observed position information according to the converted position information of each edge of the current frame and the observed position information of each edge;

[0160] According to the distance between the converted position information of each edge and the corresponding observed position information and the uncertainty of each edge, the predicted size information of the obstacle corresponding to the current frame is obtained.

[0161] In one embodiment, when the computer program is executed by a processor to implement the above-mentioned step of obtaining the observed position information of at least one edge of the current frame of the current obstacle, the uncertainty of each edge, and the predicted information of the center position according to the point cloud data and / or image data, the following steps are specifically implemented:

[0162] The point cloud data is segmented using a preset first convolutional neural network, and the segmented point cloud data is regressed using a preset second convolutional neural network to obtain prediction information of the center position and observation position information of each side;

[0163] Determine the density of the point cloud within a preset range around each edge according to the observation position information of each edge and the point cloud data;

[0164] The uncertainty of each edge is determined based on the density of the point cloud within a preset range around each edge.

[0165] In one embodiment, when the computer program is executed by a processor to implement the above-mentioned step of obtaining the predicted size information of the obstacle corresponding to the current frame according to the distance between the converted position information of each edge and the corresponding observed position information and the uncertainty of each edge, the following steps are specifically implemented:

[0166] Construct a covariance matrix based on the uncertainty on each side;

[0167] According to the covariance matrix, the distance between the converted position information of each edge and the corresponding observed position information, and the unscented Kalman filter algorithm, the predicted size information of the obstacle corresponding to the current frame is determined.

[0168] In one embodiment, when the computer program is executed by a processor, the following steps are specifically implemented:

[0169] Determine the orientation information of the current frame of the current obstacle according to the point cloud data and / or the image data;

[0170] When the computer program is executed by the processor to implement the above-mentioned step of obtaining the converted position information of each side of the current frame according to the prediction information of the center position and the current internal state data of the target filter, the following steps are specifically implemented:

[0171] According to the prediction information of the center position, the current internal state data of the target filter and the orientation information, the converted position information of each side of the current frame is obtained.

[0172] In one embodiment, when the computer program is executed by a processor to implement the above-mentioned step of determining the position information of the target rectangle according to the predicted size information and the predicted information of the center position, the following steps are specifically implemented:

[0173] The position information of the target rectangle is determined based on the predicted size information, the predicted center position information, and the orientation information.

[0174] In one embodiment, when the computer program is executed by the processor to implement the above-mentioned step of obtaining the point cloud data and / or image data of the current frame of the current obstacle, the following steps are specifically implemented:

[0175] The point cloud data and / or image data are collected by multiple sensors.

[0176] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0177] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0178] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, 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 attached claims.

Claims

1. A method for identifying the distance between a vehicle and an obstacle, characterized in that: The method comprises: Obtaining point cloud data and / or image data of a current frame of a current obstacle; The point cloud data is segmented using a preset first convolutional neural network, and the segmented point cloud data is regressed using a preset second convolutional neural network to obtain prediction information of the center position and observation position information of each side; Determining the density of the point cloud within a preset range around each of the edges according to the observed position information of each of the edges and the point cloud data; Determining the uncertainty of each edge according to the density of the point cloud within a preset range around each edge; Obtaining predicted size information of the current frame of the current obstacle according to the observed position information of each edge, the uncertainty of each edge, and the predicted information of the center position; Determining the position information of the target rectangle according to the predicted size information and the predicted information of the center position; The distance between the current obstacle and the vehicle in the current frame is determined according to the position information of the target rectangle.

2. The method according to claim 1, characterized in that The obtaining the predicted size information of the current frame of the current obstacle according to the observed position information of each edge, the uncertainty of each edge and the predicted information of the center position includes: Inputting the observed position information of each of the edges, the uncertainty of each of the edges and the predicted information of the center position into a target filter; Obtaining the converted position information of each edge of the current frame according to the prediction information of the center position and the current internal state data of the target filter, wherein the current internal state data is obtained according to the internal state data of the target filter of the previous frame and the observed position information of each edge of the current frame; Determine, according to the converted position information of each edge of the current frame and the observed position information of each edge, the distance between the converted position information of each edge and the corresponding observed position information; The predicted size information of the obstacle corresponding to the current frame is obtained according to the distance between the converted position information of each edge and the corresponding observed position information and the uncertainty of each edge.

3. The method according to claim 2, characterized in that The obtaining, according to the distance between the converted position information of each edge and the corresponding observed position information and the uncertainty of each edge, the predicted size information of the obstacle corresponding to the current frame includes: Constructing a covariance matrix based on the uncertainty of each of the edges; The predicted size information of the obstacle corresponding to the current frame is determined according to the covariance matrix, the distance between the converted position information of each edge and the corresponding observed position information, and the unscented Kalman filter algorithm.

4. The method according to claim 2, characterized in that: The method further comprises: Determining orientation information of a current frame of the current obstacle according to the point cloud data and / or the image data; The step of obtaining the converted position information of each edge of the current frame according to the prediction information of the center position and the current internal state data of the target filter comprises: The converted position information of each of the edges of the current frame is obtained according to the prediction information of the center position, the current internal state data of the target filter and the orientation information.

5. The method according to claim 4, characterized in that The step of determining the position information of the target rectangle according to the predicted size information and the predicted information of the center position includes: The position information of the target rectangle is determined based on the predicted size information, the predicted information of the center position, and the orientation information.

6. The method according to claim 1, characterized in that The step of obtaining point cloud data and / or image data of a current frame of a current obstacle includes: The point cloud data and / or image data are collected by multiple sensors.

7. A device for identifying the distance between a vehicle and an obstacle, characterized in that: The device comprises: An acquisition module, used to acquire point cloud data and / or image data of a current frame of a current obstacle; A first processing module is used to segment the point cloud data using a preset first convolutional neural network, perform circumscribed rectangle regression on the segmented point cloud data using a preset second convolutional neural network to obtain prediction information of the center position and observation position information of each edge, determine the density of the point cloud within a preset range around each edge according to the observation position information of each edge and the point cloud data, and determine the uncertainty of each edge according to the density of the point cloud within the preset range around each edge; A second processing module, configured to obtain predicted size information of the current frame of the current obstacle according to the observed position information of each edge, the uncertainty of each edge and the predicted information of the center position; A first determination module, configured to determine position information of a target rectangle according to the predicted size information and the predicted information of the center position; The second determination module is used to determine the distance between the current obstacle and the vehicle in the current frame according to the position information of the target rectangle.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images