A method for obtaining target depth information and related devices

Abstract

The present application discloses a method for obtaining target depth information and related devices. The method includes obtaining original point cloud data, obtaining first point cloud data by preprocessing the original point cloud data. The original point cloud data corresponds to original image data, and the original image data contains a 2D box of a target; obtaining all the point cloud data of the 2D box in the first point cloud data based on a preset threshold to generate second point cloud data; extracting all the point cloud data belonging to the target in the second point cloud data to generate third point cloud data; extracting the maximum and minimum point cloud data in the third point cloud data, and obtaining the depth information of the maximum and minimum point cloud data. The maximum and minimum point cloud data are the leftmost point cloud and the rightmost point cloud in the third point cloud data. The present application accurately and low-costly obtains target depth information in the case of only having a 2D box of the target through a method of fusing point cloud data, and completes the perception of object information in the world.

Description

Technical Field

[0001] This application relates to the technical field of data fusion, and in particular, to a method for obtaining target depth information and related devices. Background Art

[0002] In the real world, there are various demand scenarios that require performing operations on the collected / acquired data through relevant data processing to construct a new data set with target depth information. For example, in the fields of ADAS (Advanced Driver Assistance System) and autonomous driving, by constructing a data set with target depth information, the learning and training of the detection network for completing the target depth can be increased, and ultimately, the problem that only a monocular camera is used in the ADAS system to perceive the information of objects in the world can be solved.

[0003] However, the existing technology often highly depends on 3D boxes (such as Figure 1 A box) as input data. Generally, however, there are no foreground pixel points in the image associated with the left and right depth distances obtained from the 3D box, that is, there is no association of relevant features for the depth information in the image, and it is impossible to ensure that the deep learning network can obtain a good mapping relationship on two features with weak correlation; the prior information of the shape of the target objects such as vehicles introduced by the 3D box will also introduce a large amount of error information when the front and rear vehicles / targets are blocked (such as Figure 1 B box).

[0004] At the same time, if there is no accurate 3D box, adding new annotated 3D boxes generally increases new costs. Therefore, how to accurately and low-costly obtain target depth information is a technical problem that the existing technology needs to solve. Summary of the Invention

[0005] The technical problem to be solved by this application is to provide a method for obtaining target depth information and related devices in view of the deficiencies of the existing technology.

[0006] To solve the above technical problem, in the first aspect of the embodiments of this application, a method for obtaining target depth information is provided. The method includes:

[0007] Obtain the original point cloud data, and obtain the first point cloud data by preprocessing the original point cloud data. The original point cloud data corresponds to the original image data, and the original image data includes the 2D box of the target;

[0008] Based on a preset threshold, obtain all the point cloud data of the 2D box in the first point cloud data to generate the second point cloud data;

[0009] Extract all the point cloud data belonging to the target in the second point cloud data to generate the third point cloud data;

[0010] Extract the maximum and minimum point cloud data from the third point cloud data, and obtain the depth information of the maximum and minimum point cloud data, where the maximum and minimum point cloud data are the leftmost point cloud and the rightmost point cloud in the third point cloud data.

[0011] The method for obtaining the target depth information, wherein the preprocessing method includes: performing noise reduction processing and / or ground removal operation on the original point cloud data, the noise reduction processing is to remove the discrete point cloud of the original point cloud data, and the ground removal operation is to remove the ground point cloud data in the original point cloud data.

[0012] The method for obtaining the target depth information, wherein the ground removal operation includes:

[0013] Precision strategy, performing ground removal operation by the random sample consensus method; and / or,

[0014] Fast strategy, performing ground removal operation by a method with good far and near consistency.

[0015] The method for obtaining the target depth information, wherein obtaining all the point cloud data of the 2D box in the first point cloud data based on a preset threshold to generate a second point cloud data includes:

[0016] Project the first point cloud data into the camera coordinate system according to the joint calibration data;

[0017] Project the first point cloud data projected into the camera coordinate system into the pixel coordinate system according to the camera internal parameters;

[0018] Set the preset threshold based on the extreme values of the abscissa and ordinate of the 2D box in the pixel coordinate system;

[0019] Screen all the point cloud data in the first point cloud data in the pixel coordinate system that satisfy the preset threshold;

[0020] Generate the second point cloud data according to the screening result.

[0021] The method for obtaining the target depth information, wherein after generating the second point cloud data, it further includes projecting the second point cloud data into the camera coordinate system.

[0022] The method for obtaining the target depth information, wherein extracting all the point cloud data belonging to the target in the second point cloud data to generate a third point cloud data includes:

[0023] Set control points on the target;

[0024] Generate a number of point cloud clusters in the second point cloud data through clustering, and the clustering method includes intensity clustering and / or distance clustering;

[0025] Select a point cloud cluster containing the control points, and the selected point cloud cluster is the third point cloud data.

[0026] The method for obtaining the target depth information, wherein setting the control points on the target includes:

[0027] Select 5 points at different spatial positions from the second point cloud data according to the spatial distribution characteristics as candidate points;

[0028] Generate a number of point cloud clusters through the clustering method;

[0029] Project the point cloud cluster where the candidate points are located onto the image plane;

[0030] Generate candidate 2D boxes according to the limit values in the up, down, left, and right 4 directions of the projection result;

[0031] Calculate the IoU value of each candidate 2D box and select the candidate 2D box corresponding to the highest IoU value;

[0032] Set the candidate points of the candidate 2D box corresponding to the highest IoU value as the control points.

[0033] The second aspect of the embodiments of the present application provides an apparatus for obtaining target depth information, and the apparatus includes:

[0034] A data acquisition module, configured to acquire the original point cloud data, the corresponding original image data, and the target 2D box in the original image data.

[0035] A preprocessing module, configured to preprocess the original point cloud data to obtain the first point cloud data, wherein the preprocessing steps include performing noise reduction processing and / or ground removal operation on the original point cloud data.

[0036] An extraction module, configured to extract the second point cloud data and the third point cloud data. The extraction steps of the second point cloud data include screening all the point cloud data that meet the preset threshold in the first point cloud data in the pixel coordinate system. The extraction steps of the third point cloud data include setting control points on the target; generating a number of point cloud clusters in the second point cloud data through clustering, and the clustering method includes intensity clustering and / or distance clustering; selecting the point cloud cluster containing the control points, and the selected point cloud cluster is the third point cloud data.

[0037] A depth information acquisition module, configured to acquire the depth information of the target, and the acquisition process of the depth information includes acquiring the maximum and minimum point cloud data in the third point cloud data.

[0038] A third aspect of the embodiments of the present application provides an embedded device, characterized in that it includes: the embedded device includes the acquisition device for the target depth information as described above.

[0039] Beneficial effects: Compared with the prior art, the present application discloses a method and related device for acquiring target depth information. The method includes acquiring original point cloud data, obtaining first point cloud data by preprocessing the original point cloud data, the original point cloud data corresponding to original image data, and the original image data including a 2D box of the target; acquiring all point cloud data of the 2D box in the first point cloud data based on a preset threshold to generate second point cloud data; extracting all point cloud data belonging to the target in the second point cloud data to generate third point cloud data; extracting the maximum and minimum point cloud data in the third point cloud data, and acquiring the depth information of the maximum and minimum point cloud data, the maximum and minimum point cloud data being the leftmost point cloud and the rightmost point cloud in the third point cloud data. The present application accurately and low-costly acquires target depth information by means of point cloud data fusion in the case of only having the target 2D box, and completes the perception of object information in the world. Description of the Drawings

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative labor, other structural schematic diagrams can also be obtained according to these drawings.

[0041] Figure 1 It is a schematic diagram of the effect of introducing incorrect information by using a 3D box in the prior art.

[0042] Figure 2 It is a flowchart of a method for acquiring target depth information according to an embodiment of the present application.

[0043] Figure 3 It is a schematic diagram of the original image data.

[0044] Figure 4 It is a schematic diagram of the original point cloud data.

[0045] Figure 5 It is a schematic diagram of the principle of a method for generating third point cloud data provided by an embodiment of the present application.

[0046] Figure 6 It is a structural schematic diagram of the acquisition device for target depth information provided by the present application. Detailed Embodiments

[0047] This application provides a method for obtaining target depth information and related devices. To make the objectives, technical solutions, and effects of this application clearer and more explicit, the following further elaborates on this application with reference to the accompanying drawings and by way of examples. It should be understood that the specific examples described herein are only used to explain this application and are not used to limit this application.

[0048] Those skilled in the art of this technology can understand that, unless specifically stated otherwise, the singular forms "a", "an", "the", and "said" used herein may also include the plural forms. It should be further understood that the term "including" used in the specification of this application means the presence of the described features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or their groups. It should be understood that when we say an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used here may include wireless connection or wireless coupling. The phrase "and / or" used here includes all or any unit and all combinations of one or more related listed items.

[0049] Those skilled in the art of this technology can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used here have the same meaning as the general understanding of those of ordinary skill in the field to which this application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless specifically defined as here.

[0050] The following explains some of the terms involved in this application:

[0051] 2D box: The "2D box" described in this application refers to a two-dimensional annotation box used to annotate a target on image data. Among them, the "box" in the "2D box" should be understood as the English "bonding box", which can be interpreted as an annotation box, a bounding box, or a detection box.

[0052] 3D box: The "3D box" described in this application refers to a three-dimensional annotation box used to annotate a target on point cloud data. Among them, the "box" in the "3D box" should be understood as the English "bonding box", which can be interpreted as an annotation box, a bounding box, or a detection box.

[0053] Point cloud fusion: The "point cloud fusion" described in this application refers to the fusion of point clouds and images. In a broad sense, it can be understood as associating point cloud data and image data through a mapping relationship.

[0054] Joint calibration: It refers to obtaining the external parameters from the lidar coordinate system to the camera coordinate system through calibration. After the camera and lidar are jointly calibrated, the measurement values of the lidar can be accurately projected onto the camera image, thereby realizing the association between the laser points and the three-channel color information. Conversely, the pixels in the camera image can obtain depth values by querying the nearest laser.

[0055] It should be understood that the sequence numbers and magnitudes of the steps in this embodiment do not imply the order of execution. The order of execution of each process is determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0056] The following further describes the application content by describing the embodiments in conjunction with the accompanying drawings.

[0057] This embodiment provides a method for obtaining target depth information, as Figure 2 shown, the method includes:

[0058] S10. Obtain the original point cloud data, and obtain the first point cloud data by preprocessing the original point cloud data. The original point cloud data corresponds to the original image data, and the original image data includes the 2D box of the target.

[0059] Specifically, in this embodiment, 2 items of data need to be obtained, the original image data (such as Figure 3 ) and the original point cloud data (such as Figure 4 ), and the two are in a corresponding relationship. The original image data includes the 2D box of the target. As described in the preface, the 2D box is defined in this article as the 2D standard box of the target in the image data.

[0060] Further, in one implementation, the preprocessing method includes: performing noise reduction processing and / or ground removal operation on the original point cloud data. The noise reduction processing is to remove the discrete point cloud of the original point cloud data, and the ground removal operation is to remove the ground point cloud data in the original point cloud data.

[0061] Since the original point cloud data may have discrete points, which may lead to a decrease in the accuracy generated after the extraction of the point cloud data of the subsequent target, in one embodiment, noise reduction preprocessing is performed on the original point cloud data. Of course, this does not mean that other embodiments must also perform noise reduction preprocessing, and the result may be at the cost of a decrease in accuracy.

[0062] In addition, the preprocessing operation may also include a ground removal operation. In point cloud data, in a ground scene, if the clustering method is used for classification / segmentation, if the ground data is not removed, it will lead to a large error in the clustering effect. Therefore, it is necessary to remove the ground point cloud data. Here is an example. For the original point cloud data, if the target is a vehicle, the point cloud data within its 2D box, which is the target point cloud data, usually also includes ground point cloud data. If the ground data is not removed, when performing clustering later, when extracting the target point cloud data, not only the point cloud data of the target will be extracted, but it is very likely that all the consecutive ground point cloud data will also be extracted.

[0063] Further, the ground removal operation includes: a precise strategy, performing the ground removal operation by the random sample consensus method; and / or a fast strategy, performing the ground removal operation by a method that maintains good distance consistency.

[0064] Specifically, the applicant found through experiments that: under the condition of sufficient time, it is recommended to use the random sample consensus (Ransac method) to complete the ground removal operation, and by adjusting the parameters, an almost perfect effect of removing ground point cloud points can be achieved; if data needs to be generated quickly, it is recommended to use the fast ground removal method, which can maintain good distance consistency and avoid the situation of being misdeleted by the PCA method due to the object being too far away. At the same time, in some scenarios (such as when the road is flat), the fast ground removal method can also achieve a better ground removal effect by adjusting the vehicle height. Therefore, for the ground removal operation in the present invention, either one or a combination of the above two strategies can be selected, corresponding to 2 different ground removal methods respectively.

[0065] S20. Obtain all the point cloud data of the 2D box in the first point cloud data based on a preset threshold, and generate second point cloud data.

[0066] Further, in one implementation manner, obtaining all the point cloud data of the 2D box in the first point cloud data based on a preset threshold and generating second point cloud data specifically includes:

[0067] S21. Project the first point cloud data into the camera coordinate system according to the joint calibration data;

[0068] S22. Project the first point cloud data projected into the camera coordinate system into the pixel coordinate system according to the camera internal parameters;

[0069] S23. Set the preset threshold based on the extreme values of the abscissa and ordinate of the 2D box in the pixel coordinate system;

[0070] S24. Screen all the point cloud data that meet the preset threshold in the first point cloud data in the pixel coordinate system;

[0071] S25. Generate the second point cloud data according to the screening result.

[0072] In one embodiment, the first point cloud data is first in the radar coordinate system. At this time, the coordinate format of this point cloud can be (x l , y l , z l ). According to the joint calibration data, transfer this point cloud data from the radar coordinate system to the camera coordinate system. At this time, the coordinate format of this point cloud can be (x c , y c , z c ). Then transfer this point cloud from the camera coordinate system to the pixel coordinate system. At this time, the coordinate format of this point cloud can be (u, v, z i ). Therefore, the format of the first point cloud data at this time is (u, v, z i ). Since the original image data and the 2D box have been obtained previously, we can obtain the u min , u max , v min and v max of the 2D box, that is, set the preset threshold based on the extreme values of the abscissa and ordinate of the 2D box in the pixel coordinate system. In other words, as long as the first point cloud data simultaneously satisfies "u min ≤ u ≤ u max " and "v min ≤ v ≤ v max ", then this first point cloud data is the point cloud data within the 2D box, that is, the second point cloud data.

[0073] It should be noted that in other embodiments, the coordinate formats of the point cloud data in the radar coordinate system, camera coordinate system, and pixel coordinate system can have other expression forms, which are not limited here.

[0074] Furthermore, in one implementation, after generating the second point cloud data, it further includes re-projecting the second point cloud data back to the camera coordinate system.

[0075] The reason for re-projecting the second point cloud data back to the camera coordinate system is due to the special requirements of subsequent clustering methods in certain situations. Only when performing clustering methods on three-dimensional data can the obtained effect (accuracy) be accurate (high) enough. Since the point cloud data in the pixel coordinate system can be understood as two-dimensional data, it is necessary to re-project the second point cloud data in the pixel coordinate system back to the camera coordinate system. It should be noted that in other implementations, it may not be necessary to re-project the second point cloud data in the pixel coordinate system back to the camera coordinate system.

[0076] S30. Extract all the point cloud data belonging to the target in the second point cloud data to generate the third point cloud data.

[0077] The point cloud data within the 2D box has been presented in S20. Next, it is necessary to accurately extract the point cloud data belonging to the target from the point cloud data within the 2D box, that is, to extract the third point cloud data from the second point cloud data. The third point cloud data can be understood as the point cloud data belonging to the target.

[0078] Furthermore, in one implementation, the extraction of all the point cloud data belonging to the target from the second point cloud data to generate the third point cloud data specifically includes:

[0079] S31. Set control points on the target;

[0080] S32. Generate several point cloud clusters in the second point cloud data through clustering, and the clustering method includes intensity clustering and / or distance clustering;

[0081] S33. Select the point cloud cluster containing the control points, and the selected point cloud cluster is the third point cloud data.

[0082] Specifically, in one embodiment, as Figure 5 shown, the circumscribed box is represented as the 2D box described above. In the 2D box, there are several point cloud data. Among them, the point cloud data containing the target (the car in this embodiment) and the point cloud data outside the target (the number and the house in this embodiment) are included. Since it is physically impossible to only contain the target point cloud data within the 2D box range, there will be point cloud data outside the target within the 2D box. In this embodiment, first set a control point on the target, and then classify / split the point cloud data of different objects through the clustering method (for the clustering method, this application does not make restrictions. In one implementation, clustering can be performed through the Euclidean clustering method), and divide the point cloud data (the second point cloud data) within the 2D box into several point cloud clusters, as Figure 5 shown, divided into 3 point cloud clusters; then according to the control point of the target set previously, find the corresponding A point cloud cluster, and the A point cloud cluster is the set of target point cloud data, that is, the third point cloud data.

[0083] Furthermore, in one implementation, the setting of control points on the target specifically includes:

[0084] Select 5 points at different spatial positions as candidate points from the second point cloud data according to the spatial distribution characteristics;

[0085] Generate several point cloud clusters through the clustering method;

[0086] Project the point cloud cluster where the candidate points are located onto the image plane;

[0087] Generate candidate 2D boxes based on the extreme values of the projection results in the four directions of up, down, left, and right;

[0088] Calculate the IoU value of each of the candidate 2D boxes and select the candidate 2D box corresponding to the highest IoU value;

[0089] Set the candidate point of the candidate 2D box corresponding to the highest IoU value as the control point.

[0090] As can be seen from the above, the control point of the target is crucial. Therefore, how to select the control point is the key point of the embodiment of this application. In one embodiment, the center of the target is selected as the control point, but its effect is not accurate enough in special scenarios, resulting in inaccurate final depth information. This situation is suitable for use when there is no extremely precise requirement for depth information; in another embodiment, candidate points are selected, and then the IoU values of the 2D boxes generated by the point cloud clusters of different candidate points in the image plane are compared, and the candidate point of the 2D box with the highest IoU value is selected as the final control point. Through multiple groups of experiments, the accuracy of the finally obtained depth information is extremely high.

[0091] In other embodiments, different methods can be adopted to select the control point of the target, which will not be elaborated here.

[0092] S40. Extract the maximum and minimum point cloud data from the third point cloud data, and obtain the depth information of the maximum and minimum point cloud data, where the maximum and minimum point cloud data are the leftmost point cloud and the rightmost point cloud in the third point cloud data.

[0093] Specifically, after extracting the maximum and minimum point cloud data from the third point cloud data, it further includes projecting the third point cloud data into the pixel coordinate system.

[0094] Specifically, in one implementation manner, the extracted third point cloud data is in the camera coordinate system, and the method for obtaining the depth information includes: first projecting the third point cloud data into the pixel coordinate system, and the coordinate format of the point cloud data can be (u, v, z i ), then select the leftmost point cloud and the rightmost point cloud of the third point cloud data, and obtain the "z i " information in the above two maximum and minimum point cloud data. In another implementation manner, the method is the same as the previous implementation manner, and the difference is that finally, the distance information of the target to itself is obtained.

[0095] In summary, the present application discloses a method for obtaining target depth information and related devices. The method includes obtaining original point cloud data, obtaining first point cloud data by preprocessing the original point cloud data. The original point cloud data corresponds to original image data, and the original image data includes a 2D box of the target. Based on a preset threshold, all point cloud data of the 2D box are obtained from the first point cloud data to generate second point cloud data. All point cloud data belonging to the target are extracted from the second point cloud data to generate third point cloud data. The maximum and minimum point cloud data are extracted from the third point cloud data, and the depth information of the maximum and minimum point cloud data is obtained. The maximum and minimum point cloud data are the leftmost point cloud and the rightmost point cloud in the third point cloud data. The present application accurately and at low cost obtains target depth information in the case of only having the target 2D box through the method of point cloud data fusion, and completes the perception of object information in the world.

[0096] Based on the above method for obtaining target depth information, this embodiment provides a device for converting a target 3D box into a 2D box, as Figure 6 shown. The system includes:

[0097] A data acquisition module 100, configured to acquire original point cloud data, corresponding original image data, and a target 2D box in the original image data.

[0098] A preprocessing module 200, configured to preprocess the original point cloud data to obtain first point cloud data. Among them, the steps of the preprocessing include performing noise reduction processing and / or ground removal operation on the original point cloud data.

[0099] An extraction module 300, configured to extract second point cloud data and third point cloud data. Among them, the extraction steps of the second point cloud data include screening all point cloud data that meet the preset threshold in the first point cloud data in the pixel coordinate system. The extraction steps of the third point cloud data include setting control points on the target; generating several point cloud clusters in the second point cloud data through clustering. The clustering method includes intensity clustering and / or distance clustering; selecting the point cloud cluster containing the control points, and the selected point cloud cluster is the third point cloud data.

[0100] A depth information acquisition module 400, configured to acquire the depth information of the target. The acquisition process of the depth information includes acquiring the maximum and minimum point cloud data in the third point cloud data.

[0101] Based on the above device for obtaining target depth information, the present application further provides an embedded device. The embedded device includes the above device for obtaining target depth information to obtain target depth information through the device for obtaining target depth information.

[0102] In addition, the specific processes of loading and executing multiple instructions by the instruction processors in the above storage medium and terminal device have been described in detail in the above method, and will not be repeated here one by one.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for obtaining target depth information, characterized in that The method includes: Obtaining original point cloud data, and obtaining first point cloud data by preprocessing the original point cloud data. The original point cloud data corresponds to original image data, and the original image data contains a 2D box of a target. The preprocessing includes performing noise reduction processing and ground removal operation on the original point cloud data. The noise reduction processing is to remove the discrete point cloud of the original point cloud data, and the ground removal operation is to remove the ground point cloud data in the original point cloud data by the random sample consensus method and the method of maintaining good far and near consistency; Obtaining all the point cloud data of the 2D box in the first point cloud data based on a preset threshold, and generating second point cloud data; Selecting 5 points at different spatial positions as candidate points from the second point cloud data according to the spatial distribution characteristics, generating several point cloud clusters by a clustering method, projecting the point cloud cluster where the candidate points are located onto the image plane, generating candidate 2D boxes according to the extreme values in the up, down, left, and right 4 directions of the projection result, calculating the IoU value of each candidate 2D box, selecting the candidate 2D box corresponding to the highest IoU value, and setting the candidate points of the candidate 2D box corresponding to the highest IoU value as control points; Setting the control points on the target, generating several point cloud clusters in the second point cloud data by a clustering method. The clustering method includes intensity clustering and distance clustering, and selecting the point cloud cluster containing the control points. The selected point cloud cluster is the third point cloud data; Extracting the maximum and minimum point cloud data in the third point cloud data, and obtaining the depth information of the maximum and minimum point cloud data. The maximum and minimum point cloud data are the leftmost point cloud and the rightmost point cloud in the third point cloud data.

2. The method according to claim 1, characterized in that, The obtaining all the point cloud data of the 2D box in the first point cloud data based on a preset threshold and generating second point cloud data includes: Projecting the first point cloud data into the camera coordinate system according to the joint calibration data; Projecting the first point cloud data projected into the camera coordinate system into the pixel coordinate system according to the camera internal parameters; Setting the preset threshold based on the extreme values of the abscissa and ordinate of the 2D box in the pixel coordinate system; Screening all the point cloud data that meet the preset threshold in the first point cloud data in the pixel coordinate system; Generating the second point cloud data according to the screening result.

3. The method according to claim 1, wherein After generating the second point cloud data, it further includes projecting the second point cloud data into the camera coordinate system.

4. An apparatus for obtaining target depth information, characterized in that The device includes: A data acquisition module, configured to acquire original point cloud data, corresponding original image data, and the 2D box of the target in the original image data; A preprocessing module, configured to preprocess the original point cloud data to obtain first point cloud data. Wherein, the steps of the preprocessing include performing noise reduction processing and ground removal operation on the original point cloud data; An extraction module for extracting second point cloud data and third point cloud data. The extraction steps of the second point cloud data include screening out all point cloud data in the first point cloud data that meet a preset threshold in the pixel coordinate system. The extraction steps of the third point cloud data include selecting 5 points at different spatial positions from the second point cloud data according to the spatial distribution characteristics as candidate points, generating a number of point cloud clusters through a clustering method, projecting the point cloud cluster where the candidate points are located onto the image plane, generating candidate 2D boxes according to the extreme values in the up, down, left, and right 4 directions of the projection result, calculating the IoU value of each candidate 2D box and selecting the candidate 2D box corresponding to the highest IoU value, and setting the candidate points of the candidate 2D box corresponding to the highest IoU value as control points; setting the control points on the target, generating a number of point cloud clusters in the second point cloud data through a clustering method, the clustering method including intensity clustering and distance clustering, selecting the point cloud cluster containing the control points, and the selected point cloud cluster is the third point cloud data; A depth information acquisition module for extracting the maximum and minimum point cloud data in the third point cloud data and acquiring the depth information of the maximum and minimum point cloud data, where the maximum and minimum point cloud data are the leftmost point cloud and the rightmost point cloud in the third point cloud data.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the method for acquiring the target depth information according to any one of claims 1-3.

6. An embedded device, characterized in that, Including: The embedded device includes the device for acquiring the target depth information according to any one of claims 1-3.

Images