A dynamic object detection method, device, medium and product

Abstract

The present invention discloses a dynamic object detection method, device, medium and product, which relate to the field of autonomous environmental perception. The present invention detects dynamic objects in the environment based on a hybrid solid-state laser and a monocular vision sensor, and the detection process is as follows: first, a rough classification is performed through the semantic segmentation of the visual image, and then the image segmentation is merged based on the point cloud information of the hybrid solid-state laser to obtain a fine classification result. In addition, in order to overcome the problem that the semantics of the visual image cannot fully distinguish dynamic information, the motion posture of the hybrid solid-state laser and monocular vision component is estimated based on the laser point cloud information, and the point cloud alignment error relative to the previous moment is calculated to improve the analysis and judgment of dynamic objects in the environment, thereby realizing the accurate classification of dynamic and static objects in the environment.

Description

Technical Field

[0001] The present invention relates to the field of autonomous environmental perception, and in particular to a dynamic object detection method, equipment, medium and product. Background Art

[0002] With the progress of society and the continuous expansion of mission requirements, autonomous perception of the environment has become an essential core point in the field of intelligent control. In the civilian field, autonomous driving needs to accurately perceive road conditions and pedestrian information to facilitate control decisions. In the military field, it is necessary to determine the enemy's living forces in the environment for effective attack. Therefore, mature environmental perception methods are needed to detect dynamic objects in the scene.

[0003] Dynamic perception of environmental status is generally based on visual information, while the traditional visual frame difference method has the disadvantages of poor algorithm robustness and immobile sensors, which has strong application limitations. Deep learning can be used to judge dynamic information such as pedestrians and animals in images. However, methods based on image recognition have the problem of imprecise recognition, and methods based on image segmentation have the problem of over-segmentation. In addition, these two methods have a high missed detection rate and cannot accurately classify dynamic and static objects in the environment. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a dynamic object detection method, device, medium and product.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] In a first aspect, the present application provides a dynamic object detection method, comprising:

[0007] Acquiring sensor information; the sensor information includes laser point cloud and image information;

[0008] Segmenting the image information using a semantic segmentation algorithm to obtain a rough segmentation set; the rough segmentation set includes image regions and semantic information;

[0009] Perform sensor calibration to obtain external parameters between the hybrid solid-state laser and the monocular vision sensor and internal parameters of the monocular vision sensor;

[0010] The laser point cloud is projected into a coarse segmentation set using external parameters between a hybrid solid-state laser and a monocular vision sensor and internal parameters of the monocular vision sensor to obtain segmentation point cloud information;

[0011] Merging the segmented point cloud information to obtain a semantic result of the merged segmentation;

[0012] The laser point cloud is classified according to the semantic results of the combined segmentation to obtain a category segmentation result; the category segmentation result includes: autonomously movable point cloud segmentation, non-autonomously movable point cloud segmentation and static point cloud segmentation;

[0013] Based on the category segmentation results, a laser mileage calculation method is used to track and align the point cloud at the previous moment, and a position change of the hybrid solid-state laser and monocular vision sensor at the current moment relative to the previous moment is determined;

[0014] Determine the alignment error between the segmentation result of each category and the point cloud at the previous moment based on the posture change;

[0015] The object dynamic detection result is obtained based on the relationship between the alignment error and the set threshold.

[0016] Optionally, the laser point cloud is projected into a coarse segmentation set using external parameters between the hybrid solid-state laser and the monocular vision sensor and internal parameters of the monocular vision sensor to obtain segmentation point cloud information, specifically including:

[0017] Using external parameters between a hybrid solid-state laser and a monocular vision sensor to perform coordinate transformation on the laser point cloud to obtain three-dimensional coordinates;

[0018] Obtaining the projection of the three-dimensional coordinates on the image plane according to the internal parameters of the monocular vision sensor;

[0019] The image region where the projection of the three-dimensional coordinates on the image plane is located is checked, and the points corresponding to the projection of the three-dimensional coordinates on the image plane are associated with the image region of the coarse segmentation set to obtain the segmentation point cloud information.

[0020] Optionally, merging the segmented point cloud information to obtain a semantic result of the merged segmentation specifically includes:

[0021] In the segmented point cloud information, the edge point cloud information of adjacent image regions is extracted, and the edge point cloud information of adjacent image regions is clustered using the region growing algorithm, and the obtained clustering result is used as the semantic result of the merged segmentation.

[0022] Optionally, based on the category segmentation result, a laser mileage calculation method is used to track and align the point cloud at the previous moment, and a position change of the hybrid solid-state laser and monocular vision sensor at the current moment relative to the previous moment is determined, specifically including:

[0023] Perform point cloud sampling on the point cloud at the previous moment and the point cloud at the current moment respectively to obtain a first sampling point cloud and a second sampling point cloud;

[0024] Initializing the change in posture between the current moment and the previous moment as a unit matrix, and determining the coordinates of the second sampling point cloud based on the unit matrix;

[0025] Use the KD tree to establish a data structure of the first sampling point cloud, and search the data structure for a point that is closest to the coordinates of the second sampling point cloud as a target point;

[0026] Based on the coordinates of the target point and the coordinates of the second sampling point cloud, the unit matrix is ​​used as the iteration starting point, and the Levenberg-Marquardt algorithm is used to iteratively estimate the pose change until the set number of iterations is reached. The final iterative estimation result is used as the pose change of the hybrid solid-state laser and monocular vision sensor at the current moment relative to the previous moment.

[0027] Optionally, the determination formula of the posture change amount is:

[0028] ;

[0029] In the formula, represents the change in posture, Indicates the change in posture at the previous moment, Represents the index value of the corresponding point, represents the number of points in the second sampling point cloud, represents the point in the second sampling point cloud, Indicates the target point.

[0030] Optionally, the alignment error is determined by:

[0031] ;

[0032] In the formula, represents the alignment error, is the number of laser measurement points in the point cloud in the image area, represents the change in posture, Represents the index value of the corresponding point, represents the point in the second sampling point cloud, Indicates the target point.

[0033] Optionally, obtaining the object dynamic detection result based on the relationship between the alignment error and a set threshold value specifically includes:

[0034] When the alignment error is less than the set threshold, determining that the object dynamic detection result is a static object;

[0035] When the alignment error is greater than or equal to the set threshold, the object dynamic detection result is determined to be a dynamic object.

[0036] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described dynamic object detection methods.

[0037] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned dynamic object detection methods.

[0038] In a fourth aspect, the present application provides a computer program product, including a computer program, which implements any of the above-mentioned dynamic object detection methods when executed by a processor.

[0039] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0040] The present invention detects dynamic objects in the environment based on a hybrid solid-state laser and a monocular vision sensor, and the detection process is as follows: first, a rough classification is performed through the semantic segmentation of the visual image, and then the image segmentation is merged based on the point cloud information of the hybrid solid-state laser to obtain a fine classification result. In addition, in order to overcome the problem that the semantics of the visual image cannot fully distinguish dynamic information, the motion pose of the hybrid solid-state laser and monocular vision component is estimated based on the laser point cloud information, and the point cloud alignment error relative to the previous moment is calculated to improve the analysis and judgment of dynamic objects in the environment, thereby realizing the accurate classification of dynamic and static objects in the environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 This is a flow chart of a dynamic object detection method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] The purpose of the present invention is to provide a dynamic object detection method, device, medium and product, aiming to overcome the problem of incomplete distinction of dynamic information by visual image semantics, to improve the analysis and judgment of dynamic objects in the environment, and at the same time, to accurately classify dynamic and static objects in the environment.

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] Example 1

[0047] Hybrid solid-state lasers (such as DJI Livox AVIA lasers, etc.) have the characteristics of high field of view coverage and strong working stability. They can directly obtain the distance information of the environment and can complement the information with the visual sensor. Based on this, the present embodiment is configured to detect dynamic objects in the environment based on a hybrid solid-state laser-monocular vision component. Specifically, a rough classification is first performed through the semantic segmentation of the visual image, and then the image segmentation is merged based on the point cloud information of the hybrid solid-state laser to obtain a refined classification result. In addition, in order to overcome the problem that the semantics of the visual image cannot fully distinguish dynamic information, the motion pose estimation of the hybrid solid-state laser and monocular vision component is performed based on the laser point cloud information, and the point cloud alignment error relative to the previous moment is calculated to improve the analysis and judgment of dynamic objects in the environment.

[0048] Based on the above description, if Figure 1 As shown, a dynamic object detection method provided in this embodiment includes:

[0049] Step 100: Acquire sensor information. The sensor information is obtained by hybrid solid-state laser and monocular vision sensor; the sensor information includes laser point cloud and image information. Assume that the current time is , by collecting information from a hybrid solid-state laser and monocular vision sensor, the laser point cloud is obtained as follows: , the image information is .

[0050] Step 101: Use a semantic segmentation algorithm to segment the image information and obtain a rough segmentation set. The rough segmentation set includes image regions and semantic information. The semantic segmentation algorithm used can be ParseNet, SegNet, Segmenter algorithm, etc. ParseNet, SegNet, and Segmenter algorithms are all existing image semantic segmentation algorithms based on deep neural networks. By inputting image information into the above image segmentation algorithms, the image segmentation result can be obtained.

[0051] For example, semantic segmentation algorithms can be used to analyze image information. Segmentation is performed to obtain a rough segmentation set, namely ,in, is the first A segmentation result, which includes the segmented image area and semantic information.

[0052] Step 102: perform sensor calibration to obtain external parameters between the hybrid solid-state laser and the monocular vision sensor and internal parameters of the monocular vision sensor. The external parameters are recorded as , which can contain 3D translation parameters and the three-dimensional rotation parameters The internal parameters are recorded as .

[0053] Step 103: Project the laser point cloud into the coarse segmentation set using the external parameters between the hybrid solid-state laser and the monocular vision sensor and the internal parameters of the monocular vision sensor to obtain segmentation point cloud information.

[0054] In practical applications, laser point cloud Middle Points Take the projection process in this step as an example, as follows:

[0055] Step 1, coordinate system conversion: Use the external parameters between the hybrid solid-state laser and the monocular vision sensor to transform the laser point cloud in the laser coordinate system Middle Points Convert to the visual coordinate system to get the three-dimensional coordinates ,Right now .

[0056] Step 2: Visual projection: Based on the internal parameters of the monocular vision sensor , get the three-dimensional coordinates The projection on the image plane is denoted by .in, is the three-dimensional coordinate The depth value of .

[0057] Step 3: Check the projection The image area where the projection is located (i.e. the coarse segmentation area) The corresponding point Associated to the coarse segmentation area. Complete the laser point cloud After associating all the points in All have associated point clouds In actual application, the inspection here refers to: Check In which coarse segmentation area of ​​the image information.

[0058] Step 104: Merge the segmented point cloud information to obtain the semantic result of the merged segmentation. In order to solve the problem of over-segmentation in visual images, the segmentation is merged based on the point cloud information corresponding to the rough segmentation of the image to obtain a more accurate segmentation result. In order to reduce the amount of calculation for segmentation merging, only the edge point cloud information of adjacent image areas is extracted from the segmented point cloud information, and the edge point cloud information of adjacent image areas (i.e., the rough segmentation area) is clustered using the region growing algorithm, and the obtained clustering result is used as the semantic result of the merged segmentation. The specific process is as follows:

[0059] Step 1: Extract the edge point cloud of the rough visual segmentation area. Define the distance to the segmentation edge in the visual segmentation. The area within pixels is the visual segmentation edge area. Therefore, according to the projection relationship from point cloud to image information in step 103, the point cloud corresponding to the visual segmentation edge area can be obtained. As an example, assume that its adjacent visual segmentation set is . Let the segmented image region The corresponding visual segmentation edge area point cloud is , and its point cloud set near the edge area of ​​visual segmentation is .

[0060] Step 2: Clustering of visual segmentation edge region point clouds. According to the visual segmentation edge region point clouds and their adjacent visual segmentation edge region point clouds, we can obtain The adjacent point cloud of each adjacent (i.e., adjacent) visual segmentation edge area point cloud, i.e., the set .in, Above " indicates the adjacency of the point cloud, and the brackets in the set Indicates the relationship between adjacent. For example, use the region growing method for clustering:

[0061] Assumptions The number of midpoints is , traverse At each point in Find the point with the closest distance to it. Calculate The average of the closest distances is ,if If it is less than the set threshold, it means If the two segments in belong to the same class, Its adjacent visual segmentation area Perform clustering and merging, and set the semantics of the merged segmentation to the semantic information of the larger area in the adjacent segmentation. Merge with other adjacent visual segmentations and update the associated point cloud corresponding to the merged segmentation The result after segmentation and merging is set to .

[0062] Step 3: Based on the results of segmentation and merging , continue to merge using Step 1 and Step 2 until the merge cannot be continued (for example, the previous Step 2 is greater than or equal to the set threshold). Let the final segmentation and merging result be , so far the segmentation point cloud information is obtained.

[0063] Step 105: Classify the laser point cloud according to the semantic results of the merged segmentation to obtain the category segmentation results. The category segmentation results include: autonomously movable point cloud segmentation, non-autonomous movable point cloud segmentation and static point cloud segmentation. The classification method of each category is as follows: the point cloud with semantic information of dynamic objects such as pedestrians and animals is determined as autonomously movable point cloud segmentation. The point cloud with semantic information of static objects such as the ground and buildings is determined as static point cloud segmentation. The point cloud corresponding to the remaining semantic information such as vehicles, ships, and aircraft is the non-autonomous movable point cloud segmentation.

[0064] Step 106: Based on the category segmentation result, the laser mileage calculation method is used to track and align the point cloud at the previous moment, and the position change of the hybrid solid-state laser and monocular vision sensor at the current moment relative to the previous moment is determined.

[0065] In the actual application process, in order to further judge the dynamic properties of the non-autonomous moving point cloud segmentation, the hypothesis test method is used to determine its true motion state. For example, assuming that the non-autonomous moving point cloud is in a static state, according to the non-autonomous moving point cloud segmentation and static point cloud segmentation, the laser mileage calculation method is used to calculate the last moment (i.e. Time) point cloud is tracked and aligned, and the current Hybrid solid-state laser-monocular vision components relative to The change of posture at the moment Among them, the laser mileage calculation method is designed as follows:

[0066] Step 1. Point cloud downsampling: Assume The point cloud at time , The point cloud at time In order to reduce the amount of subsequent calculations, Point cloud at a moment and Point cloud at a moment Sample one point every three points, and the point cloud obtained after sampling is the first sampling point cloud and the second sampling point cloud .

[0067] Step 2: Establish the corresponding relationship: Use KD tree to establish The downsampled point cloud at time (i.e. the first sampling point cloud ) data structure. Moment and The change of posture at the moment Initialized to the identity matrix, the second sampling point cloud Points in exist The coordinates in the time coordinate system are Through the KD tree Find and The closest point is set to Therefore, we get and The correspondence of the midpoints.

[0068] Step 3, pose estimation: based on pose change The initial value of the identity matrix is ​​used as the starting point of the iteration, and the Levenberg-Marquardt algorithm is used to calculate the pose change Iterative estimation of , that is:

[0069] .

[0070] in, Represents the index value of the corresponding point, express The number of midpoints.

[0071] Before subsequent iterations, update according to the estimated pose change , and re-establish through Step2 and The corresponding relationship between the midpoints is obtained, and then the iterative estimation process of Step 3 is performed. When the number of iterations reaches the set value, or the error in the iteration process When the decrease rate of is less than the set value, the final estimate of the posture change is obtained, which is recorded as .

[0072] Step 107: Determine the alignment error between the segmentation results of each category and the point cloud at the previous moment based on the change in posture. Image area for non-autonomous moving point cloud segmentation For example, the alignment error is calculated as:

[0073] .

[0074] in, For the segmented point cloud The number of laser measurement points in .

[0075] Step 108: Obtain the object dynamic detection result based on the relationship between the alignment error and the set threshold. When the alignment error is less than the set threshold, the object dynamic detection result is determined to be a static object. When the alignment error is greater than or equal to the set threshold, the object dynamic detection result is determined to be a dynamic object.

[0076] In actual application, based on the assumption in step 106, if the alignment error is less than the set threshold, it indicates that Image region segmented at a certain moment Relative to If there is no movement at the moment, the point cloud that cannot be moved autonomously is determined to be segmented as a static object, otherwise it is a dynamic object.

[0077] In summary, the determination of dynamic objects in the environment includes the segmentation of autonomously movable point clouds and the segmentation of non-autonomously movable point clouds that are determined to be dynamic objects. At the same time, the corresponding dynamic area of ​​the visual image can also be obtained based on the correspondence between the point cloud segmentation and the visual image.

[0078] Example 2

[0079] A computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the dynamic object detection method in Embodiment 1.

[0080] Example 3

[0081] A computer-readable storage medium stores a computer program, which implements the dynamic object detection method in embodiment 1 when executed by a processor.

[0082] Example 4

[0083] A computer program product includes a computer program, and when the computer program is executed by a processor, the dynamic object detection method in embodiment 1 is implemented.

[0084] Example 5

[0085] A computer device, which may be a database. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. 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 pending transactions. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the dynamic object detection method in Example 1 can be implemented.

[0086] 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 the memory, database or other medium used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided by the present invention may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided by the present invention may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0087] The technical features of the above embodiments may be arbitrarily combined. 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.

[0088] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core ideas of the present invention. The same and similar parts between the embodiments can be referred to each other. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A dynamic object detection method, characterized in that: The method comprises: Step 100: Acquire sensor information; the sensor information is obtained by a hybrid solid-state laser and a monocular vision sensor; the sensor information includes a laser point cloud and image information; Step 101: Segment the image information using a semantic segmentation algorithm to obtain a coarse segmentation set; the coarse segmentation set includes image regions and semantic information; Step 102: performing sensor calibration to obtain external parameters between the hybrid solid-state laser and the monocular vision sensor and internal parameters of the monocular vision sensor; Step 103: Projecting the laser point cloud into a coarse segmentation set using external parameters between the hybrid solid-state laser and the monocular vision sensor and internal parameters of the monocular vision sensor to obtain segmentation point cloud information; Step 104: merging the segmented point cloud information to obtain a semantic result of the merged segmentation, specifically including: In the segmented point cloud information, edge point cloud information of adjacent image regions is extracted, and the edge point cloud information of adjacent image regions is clustered using a region growing algorithm, and the obtained clustering result is used as the semantic result of the merged segmentation; Step 105: classify the laser point cloud according to the semantic results of the combined segmentation to obtain a category segmentation result; the category segmentation result includes: autonomously movable point cloud segmentation, non-autonomously movable point cloud segmentation and static point cloud segmentation; Step 106: Based on the category segmentation result, the laser mileage calculation method is used to track and align the point cloud at the previous moment, and the position change of the hybrid solid-state laser and monocular vision sensor at the current moment relative to the previous moment is determined, specifically including: Perform point cloud sampling on the point cloud at the previous moment and the point cloud at the current moment respectively to obtain a first sampling point cloud and a second sampling point cloud; Initializing the change in posture between the current moment and the previous moment as a unit matrix, and determining the coordinates of the second sampling point cloud based on the unit matrix; Use the KD tree to establish a data structure of the first sampling point cloud, and search the data structure for a point that is closest to the coordinates of the second sampling point cloud as a target point; Based on the coordinates of the target point and the coordinates of the second sampling point cloud, the unit matrix is ​​used as the iteration starting point, and the Levenberg-Marquardt algorithm is used to iteratively estimate the pose change amount until the set number of iterations is reached, and the final iterative estimation result is used as the pose change amount of the hybrid solid-state laser and monocular vision sensor at the current moment relative to the previous moment; the determination formula of the pose change amount is: ； In the formula, represents the change in posture, Indicates the change in posture at the previous moment, Represents the index value of the corresponding point, represents the number of points in the second sampling point cloud, represents the point in the second sampling point cloud, Indicates the target point; Step 107: Determine the alignment error between the segmentation results of each category and the point cloud at the previous moment based on the change in posture; the formula for determining the alignment error is: ； In the formula, represents the alignment error, is the number of laser measurement points in the point cloud in the image area, represents the change in posture, Represents the index value of the corresponding point, represents the point in the second sampling point cloud, Indicates the target point; Step 108: Obtaining the object dynamic detection result based on the relationship between the alignment error and the set threshold, specifically including: When the alignment error is less than the set threshold, determining that the object dynamic detection result is a static object; When the alignment error is greater than or equal to the set threshold, the object dynamic detection result is determined to be a dynamic object.

2. The dynamic object detection method according to claim 1, characterized in that: The laser point cloud is projected onto a coarse segmentation set using external parameters between the hybrid solid-state laser and the monocular vision sensor and internal parameters of the monocular vision sensor to obtain segmentation point cloud information, specifically including: Using external parameters between a hybrid solid-state laser and a monocular vision sensor to perform coordinate transformation on the laser point cloud to obtain three-dimensional coordinates; Obtaining the projection of the three-dimensional coordinates on the image plane according to the internal parameters of the monocular vision sensor; The image region where the projection of the three-dimensional coordinates on the image plane is located is checked, and the points corresponding to the projection of the three-dimensional coordinates on the image plane are associated with the image region of the coarse segmentation set to obtain the segmentation point cloud information.

3. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the dynamic object detection method according to any one of claims 1 to 2.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the dynamic object detection method according to any one of claims 1 to 2 is implemented.

5. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the dynamic object detection method according to any one of claims 1 to 2 is implemented.

Images