Three-dimensional dynamic object recognition methods, systems and storage media

Abstract

This invention discloses a method, system, and storage medium for recognizing dynamic objects in three-dimensional space. Applied to the field of object recognition technology, it can accurately identify dynamic objects in three-dimensional space and effectively improve system stability. The method includes: acquiring point cloud data of a target three-dimensional space; voxelizing the target three-dimensional space to obtain several spatial voxels; calculating the voxel value of each spatial voxel in the current point cloud frame based on the point cloud data; wherein the voxel value is determined based on the point cloud data in the spatial voxel; updating the cumulative state data of each spatial voxel based on the voxel value; converting the cumulative state data into voxel motion coefficients using a preset motion indication function; wherein the preset motion indication function includes a nonlinear function; and recognizing dynamic objects in the target three-dimensional space based on the voxel motion coefficients.

Description

Technical Field

[0001] This invention relates to the field of object recognition technology, and in particular to a method, system and storage medium for three-dimensional dynamic object recognition. Background Technology

[0002] In computer vision, the motion of objects is viewed as a change in the environment, and the detection of moving objects in space essentially involves identifying these changes. The recognition of spatial object states, as a crucial infrastructure layer, has extensive practical value. Related technologies primarily rely on background image verification or target recognition to identify moving objects. However, the untimely updating of background images can lead to misjudgments when using background image verification, resulting in insufficient real-time performance and stability. Target recognition, on the other hand, can only identify training targets, limiting its scope and making it susceptible to weather conditions, hindering all-weather recognition. Summary of the Invention

[0003] To address at least one of the aforementioned technical problems, this invention proposes a method, system, and storage medium for recognizing dynamic objects in three-dimensional space, which can accurately identify dynamic objects in three-dimensional space and effectively improve the stability of the system.

[0004] On one hand, embodiments of the present invention provide a method for recognizing dynamic objects in three-dimensional space, including the following steps:

[0005] Acquire point cloud data of the target in three-dimensional space;

[0006] The target three-dimensional space is voxelized to obtain several spatial voxels;

[0007] The voxel value of each spatial voxel in the current point cloud frame is calculated based on the point cloud data; wherein the voxel value is determined based on the point cloud data in the spatial voxel;

[0008] Update the cumulative state data of each spatial voxel based on the voxel value;

[0009] The accumulated state data is converted into voxel motion coefficients using a preset motion indication function; wherein, the preset motion indication function includes a nonlinear function;

[0010] The dynamic objects in the target three-dimensional space are identified based on the voxel motion coefficients.

[0011] A method for recognizing dynamic objects in three-dimensional space according to an embodiment of the present invention has at least the following beneficial effects: First, this embodiment acquires point cloud data of the target three-dimensional space and voxels the target three-dimensional space to obtain several spatial voxels. Then, this embodiment calculates the voxel value of each spatial voxel in the current point cloud frame based on the point cloud data, and updates the cumulative state data of each spatial voxel based on the voxel value. Next, the cumulative state data of each spatial voxel is converted into voxel motion coefficients through a preset motion indication function, thereby recognizing dynamic objects in the target three-dimensional space based on the voxel motion coefficients, achieving relatively accurate recognition of dynamic objects in three-dimensional space. Simultaneously, this embodiment, by acquiring point cloud data in the target three-dimensional space to recognize dynamic objects, can effectively alleviate the problem of images being greatly affected by the environment, and does not rely on a background image for recognition, alleviating the problems of background image maintenance and real-time performance, and effectively improving the stability of the system.

[0012] According to some embodiments of the present invention, acquiring point cloud data of the target three-dimensional space includes:

[0013] A preset point cloud acquisition device is set in the target three-dimensional space; the preset point cloud acquisition device includes a lidar, a structured light scanning device, and a binocular stereo vision device.

[0014] The point cloud data of the target three-dimensional space is acquired in real time by the preset point cloud acquisition device.

[0015] According to some embodiments of the present invention, the step of voxelizing the target three-dimensional space to obtain a plurality of spatial voxels includes:

[0016] The voxel size data of the spatial voxel is determined according to preset voxel conditions; wherein, the preset voxel conditions include preset error tolerance, system computing power, and resolution of the point cloud data;

[0017] The target three-dimensional space is voxelized based on the voxel size data to obtain the plurality of spatial voxels.

[0018] According to some embodiments of the present invention, calculating the voxel value of each spatial voxel in the current point cloud frame based on the point cloud data includes:

[0019] The distribution data of the point cloud in each spatial voxel in the current point cloud frame is determined based on the point cloud data.

[0020] Based on the distribution data, it is determined that there are cloud points within the first spatial voxel, and the voxel value of the first spatial voxel is set to the active state.

[0021] Based on the distribution data, it is determined that there are no cloud points within the second spatial voxel, and the voxel value of the second spatial voxel is set to a static state.

[0022] According to some embodiments of the present invention, updating the cumulative state data of each spatial voxel based on the voxel value includes:

[0023] The voxel value of the spatial voxel is determined to be the active state, and the cumulative data of the state of the spatial voxel is accumulated.

[0024] Alternatively, the voxel value of the spatial voxel is determined to be the static state, and the accumulated state data of the spatial voxel is attenuated.

[0025] According to some embodiments of the present invention, after performing the step of identifying the dynamic object in the target three-dimensional space based on the voxel motion coefficients, the method further includes:

[0026] A four-dimensional point cloud is constructed based on the voxel motion coefficients and the point cloud data;

[0027] The four-dimensional point cloud is input into a preset clustering algorithm for cluster analysis to obtain the three-dimensional data of the dynamic object; wherein, the three-dimensional data includes the length data, width data and height data of the dynamic object.

[0028] According to some embodiments of the present invention, the preset clustering algorithm includes a density clustering algorithm;

[0029] The step of inputting the four-dimensional point cloud into a preset clustering algorithm for cluster analysis to obtain the three-dimensional data of the dynamic object includes:

[0030] The three-dimensional data is obtained by performing cluster analysis on the four-dimensional point cloud using the density clustering algorithm.

[0031] On the other hand, embodiments of the present invention also provide a three-dimensional spatial dynamic object recognition system, including:

[0032] The acquisition module is used to acquire point cloud data of the target's three-dimensional space;

[0033] A voxelization module is used to voxelize the target three-dimensional space to obtain several spatial voxels;

[0034] The calculation module is used to calculate the voxel value of each spatial voxel in the current point cloud frame based on the point cloud data; wherein the voxel value is determined based on the point cloud data in the spatial voxel;

[0035] The update module is used to update the cumulative state data of each spatial voxel based on the voxel value;

[0036] A conversion module is used to convert the accumulated state data into voxel motion coefficients using a preset motion indication function; wherein the preset motion indication function includes a nonlinear function.

[0037] The identification module is used to identify dynamic objects in the target three-dimensional space based on the voxel motion coefficients.

[0038] On the other hand, embodiments of the present invention also provide a three-dimensional spatial dynamic object recognition system, including:

[0039] At least one processor;

[0040] At least one memory for storing at least one program;

[0041] When the at least one program is executed by the at least one processor, the at least one processor implements the three-dimensional dynamic object recognition method as described in the above embodiments.

[0042] On the other hand, embodiments of the present invention also provide a computer storage medium storing a processor-executable program, which, when executed by the processor, is used to implement the three-dimensional spatial dynamic object recognition method as described in the above embodiments. Attached Figure Description

[0043] Figure 1 This is a flowchart of the three-dimensional dynamic object recognition method provided in the embodiments of the present invention;

[0044] Figure 2 This is a block diagram illustrating the principle of a three-dimensional dynamic object recognition system provided in an embodiment of the present invention. Detailed Implementation

[0045] The embodiments described in this application should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0046] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0048] In computer vision, motion is typically viewed as a change in the environment, and the detection of moving objects in space is essentially about recognizing this environmental change. The recognition of spatial object states, as a high-level infrastructure, has extensive practical value. In related technologies, moving object recognition usually requires storing a background image and filtering moving objects by comparing it with the background image. However, background image-based verification schemes suffer from increased costs due to the need for regular maintenance of the background image, and the update frequency of the background image also needs to be considered, which is related to downstream business processes, introducing various variables. Furthermore, if the background image is not updated in a timely manner when the detection environment changes, misjudgments can occur, significantly impacting the accuracy and stability of the recognition results. Additionally, for dynamic object recognition relying on target recognition methods, it can only identify pre-trained targets, failing to recognize untrained targets or determine motion, which greatly limits the application scenarios of this method. Furthermore, image-based target recognition methods are greatly affected by weather conditions. For example, in poor lighting conditions or rainy or foggy weather, it is difficult to achieve all-weather recognition, which greatly affects the accuracy of recognizing the motion state of objects.

[0049] Based on this, one embodiment of the present invention provides a method, system, and storage medium for recognizing dynamic objects in three-dimensional space, which can accurately identify dynamic objects in three-dimensional space and effectively improve the stability of the system. (Refer to...) Figure 1 The method in this embodiment of the invention includes, but is not limited to, steps S110, S120, S130, S140, S150 and S160.

[0050] Specifically, the application process of the method in this embodiment of the invention includes, but is not limited to, the following steps:

[0051] S110: Acquire point cloud data of the target's three-dimensional space.

[0052] S120: Voxelize the target three-dimensional space to obtain several spatial voxels.

[0053] S130: Calculate the voxel value of each spatial voxel in the current point cloud frame based on the point cloud data. The voxel value is determined based on the point cloud data in the spatial voxel.

[0054] S140: Update the cumulative state data of each spatial voxel based on the voxel value.

[0055] S150: Converts accumulated state data into voxel motion coefficients using a preset motion indicator function. The preset motion indicator function includes a nonlinear function.

[0056] S160: Identify dynamic objects in the target's three-dimensional space based on voxel motion coefficients.

[0057] In this specific embodiment, the first step is to acquire point cloud data of the target 3D space. This embodiment obtains the point cloud data of the target 3D space in real-time by scanning the target 3D space. Then, this embodiment voxels the target 3D space to obtain several spatial voxels. Specifically, a voxel is a combination of pixel, volume, and element, equivalent to a pixel in 3D space, and this pixel is three-dimensional. Treating the entire target 3D space as composed of spatial voxels, each region in the target 3D space corresponds to a specific voxel. Further, this embodiment calculates the voxel value of each spatial voxel in the current point cloud frame based on the point cloud data. Specifically, the voxel value of each spatial voxel is determined based on the point cloud data within the spatial voxel. For example, this embodiment obtains the current point cloud frame by acquiring the point cloud data of the target 3D space in real-time. Then, the voxel value of each spatial voxel in the current point cloud frame is determined based on the point cloud data of the current point cloud frame. Next, this embodiment updates the cumulative state data of each spatial voxel based on the calculated voxel values ​​of each spatial voxel in the current point cloud frame. In this embodiment, each spatial voxel has corresponding cumulative state data to statistically analyze the changes in voxel values. Further, this embodiment converts the cumulative state data into voxel motion coefficients using a preset motion indicator function. Specifically, the preset motion indicator function in this embodiment includes a nonlinear function. This embodiment expresses the motion state of objects in space using voxel motion coefficients, thereby increasing the expressive power of the detection scene by converting the cumulative state data into voxel motion coefficients using the preset motion indicator function. Since the cumulative state data is based on statistical data, i.e., linear data, this embodiment converts the cumulative state data into voxel motion coefficients using some nonlinear motion state indicator functions to enhance the nonlinear expression of the target three-dimensional space. The selection of the nonlinear function in this embodiment is set according to the downstream business requirements. For example, when it is necessary to use voxel motion coefficients to determine the probability that an object in the region corresponding to a spatial voxel is stationary, a function that maps the cumulative state data to the [0,1] interval, such as the sigmoid function, can be selected. When it is necessary to filter out objects that are absolutely stationary or have a very high probability of being stationary in the target three-dimensional space scene, the preset motion indicator function can be set as the selection function, as shown in the following equation (1):

[0058]

[0059] In the formula, x represents the cumulative state data, and y represents the voxel motion coefficient.

[0060] Furthermore, this embodiment identifies dynamic objects in the target 3D space based on voxel motion coefficients. This embodiment converts the accumulated state data into voxel motion coefficients using a preset motion indicator function, then analyzes the voxel motion coefficients of each spatial voxel to filter out dynamic objects in the target 3D space. For example, the preset motion indicator function is the selection function shown in equation (1) above. When x is greater than 90, the object corresponding to its spatial voxel is likely a stationary object, and its corresponding motion coefficient is 1; otherwise, it is 0, thus distinguishing between dynamic and static objects in the target 3D space. This embodiment identifies dynamic objects in 3D space by acquiring point cloud data of the target 3D space, effectively alleviating the problem of images being greatly affected by the environment. Furthermore, this embodiment does not rely on a background image for identification, effectively alleviating the problems of background image maintenance and real-time performance, and improving the stability and reliability of the system.

[0061] In some embodiments of the present invention, point cloud data of the target three-dimensional space is acquired, including but not limited to:

[0062] A preset point cloud acquisition device is set up in the target's three-dimensional space. The preset point cloud acquisition device includes a lidar, a structured light scanning device, and a binocular stereo vision device.

[0063] Point cloud data of the target three-dimensional space is acquired in real time through a preset point cloud acquisition device.

[0064] In this specific embodiment, a preset point cloud acquisition device is set up in the target three-dimensional space to acquire point cloud data of the target three-dimensional space in real time. Specifically, the preset point cloud acquisition device in this embodiment includes a LiDAR, a structured light scanning device, and a binocular stereo vision device. Different preset point cloud acquisition devices are selected according to different target three-dimensional space scenes. For example, LiDAR has good overall performance but is relatively expensive; therefore, when cost is not a concern, LiDAR can be used to acquire point cloud data. Structured light scanning devices are greatly affected by lighting conditions, performing poorly in strong light and having a short detection distance, but they perform well in low-light environments. Therefore, in application scenarios that require adaptation to low-light environments and do not have high distance requirements, structured light scanning devices can be used to acquire point cloud data. Meanwhile, binocular stereo vision devices are based on ordinary cameras and are relatively inexpensive, but they perform poorly in dim scenes and scenes with weak textures. Therefore, when the lighting conditions of the measurement environment are good and the scene texture is rich, binocular stereo vision devices can be used to acquire point cloud data. Furthermore, when setting up a preset point cloud acquisition device in the target 3D space, it is necessary to consider the performance of the hardware, specifically the performance of the preset point cloud acquisition device. For example, when the preset point cloud acquisition device is a LiDAR, based on the relationship between the area to be detected and the resolution, the point cloud density should be increased as much as possible while completely covering the detection area to improve the subsequent recognition accuracy. Simultaneously, the stability of the device must be ensured during deployment. It is easy to understand that if the preset point cloud acquisition device is unstable, it can easily lead to large errors in the recognition results, affecting the accuracy of recognition. Furthermore, the stability of the device and the industrial control computer network must be guaranteed to ensure that the real-time point cloud data acquired by the preset point cloud acquisition device can be transmitted in a timely manner.

[0065] In some embodiments of the present invention, the target three-dimensional space is voxelized to obtain several spatial voxels, including but not limited to:

[0066] The voxel size data of spatial voxels are determined based on preset voxel conditions. These preset voxel conditions include preset error tolerance, system computing power, and point cloud data resolution.

[0067] Based on the voxel size data, the target three-dimensional space is voxelized to obtain several spatial voxels.

[0068] In this specific embodiment, the voxel size data of the spatial voxels is first determined according to preset voxel conditions. Then, the target 3D space is voxelized according to the voxel size data to obtain several spatial voxels. Specifically, the preset voxel conditions include preset error tolerance, system computing power, and point cloud data resolution. This embodiment determines the voxel size data of the spatial voxels by combining the preset tolerance, system computing power, and point cloud data resolution, thereby voxelizing the target 3D space to obtain several spatial voxels. A voxel can be understood as a 3D raster map with a fixed resolution, that is, the target 3D space is considered to be composed of several spatial voxels. Each region in the target 3D space corresponds to a corresponding voxel value. In this embodiment, the process of voxelizing the target 3D space is equivalent to the process of modeling the target 3D space. After setting up the preset point cloud acquisition device, this embodiment uses the location of the preset point cloud acquisition device as the origin of the voxel space to reduce the computational load in calculating the relationship between the point cloud and the spatial voxels. Meanwhile, the origin of the voxel space can also be customized according to the needs of downstream applications. In this case, it is necessary to first measure and calculate the rotation and translation matrix of the preset point cloud acquisition device to the set voxel origin, and then use this matrix to calculate the correspondence between the point cloud and the spatial voxels. Further, in this embodiment, the voxel size of the spatial voxels is determined according to the preset error tolerance, system computing power, and resolution of the point cloud data. Specifically, in this embodiment, within the system's computing power and resolution tolerance range, the voxel size data is made as small as possible to improve the accuracy and precision of recognition. In this embodiment, when the point cloud data resolution is too low, if the number of spatial voxels is small, it is easy for the spatial voxels corresponding to stationary objects to accumulate enough state accumulation data, making it difficult to distinguish between moving and stationary objects. It should be noted that for voxel size data, the set number of spatial voxels in a unit space is less than half of the number of point clouds falling in that unit space.

[0069] Furthermore, this embodiment needs to balance real-time performance and accuracy. For example, when high real-time performance is required, the frame rate needs to be increased, which increases the voxel size due to hardware limitations. Conversely, when real-time performance is less critical but accuracy is more important, the frame rate and voxel size can be reduced. When downstream applications have lower real-time performance requirements but higher accuracy requirements, this embodiment performs frame extraction on the point cloud to reduce the computational power saved by frame extraction. For example, if the preset frame rate of the point cloud acquisition device is 10 frames per second, without frame extraction, 10 frames of point cloud data need to be processed per second. When hardware performance is insufficient or the application layer has a lower frame rate requirement (i.e., 10 frames per second is not necessary), this embodiment reduces the computational load by performing frame extraction. This embodiment processes the acquired 10 frames of point cloud data every preset frame and discards unprocessed frames, thus achieving frame extraction of the point cloud data.

[0070] In some embodiments of the present invention, the voxel values ​​of each spatial voxel in the current point cloud frame are calculated based on point cloud data, including but not limited to:

[0071] The distribution data of point cloud in each spatial voxel in the current point cloud frame is determined based on the point cloud data.

[0072] Based on the distribution data, cloud points are determined to exist within the first spatial voxel, and the voxel value of the first spatial voxel is set to the active state.

[0073] Based on the distribution data, it is determined that there are no cloud points within the second space voxel, and the voxel value of the second space voxel is set to a static state.

[0074] In this specific embodiment, the distribution data of the point cloud in each spatial voxel in the current point cloud frame is first determined based on the acquired point cloud data. Then, based on the data, it is determined whether cloud points exist in each spatial voxel, and the voxel value of the corresponding spatial voxel is set. Specifically, when it is determined that a point cloud is distributed in a certain spatial voxel (i.e., the first spatial voxel) based on the distribution data of the point cloud in the current point cloud frame, the voxel value of the first spatial voxel is set to an active state. It is easy to understand that in this embodiment, if a cloud point in the current point cloud frame falls within a certain spatial voxel, that spatial voxel is activated, i.e., its voxel value is set to an active state. Correspondingly, when it is determined that a cloud point is not present in a certain spatial voxel (i.e., the second spatial voxel) based on the distribution data, the voxel value of the second spatial voxel is set to a static state. In this embodiment, each spatial voxel is divided into a first spatial voxel and a second spatial voxel, and the first and second spatial voxels are classified according to whether cloud points exist in the spatial voxel. Furthermore, since the number of point clouds falling on spatial voxels is related to resolution, and the distribution at each location in the same frame of point cloud is uneven, while resolution and point cloud distribution are not related to whether the object is in motion, this embodiment determines the voxel value of each spatial voxel by judging whether there are any points falling within a spatial voxel in the current point cloud frame. For example, if there are cloud points falling within a spatial voxel in the current point cloud frame, the voxel value of that spatial voxel is set to 1, i.e., in an active state. Conversely, if no cloud points fall within a spatial voxel, the voxel value of that spatial voxel is set to 0, i.e., in a stationary state.

[0075] In some embodiments of the present invention, the cumulative state data of each spatial voxel is updated according to the voxel value, including but not limited to:

[0076] The voxel value of the spatial voxel is determined to be in the active state, and the state accumulation data of the spatial voxel is accumulated.

[0077] Alternatively, determine that the voxel value of the spatial voxel is in a static state, and then decay the accumulated state data of the spatial voxel.

[0078] In this specific embodiment, the accumulated state data of each spatial voxel is updated through an accumulation and decay mechanism. Specifically, this embodiment determines the voxel value of the spatial voxel. When the voxel value is active, the accumulated state data of that spatial voxel is accumulated; when the voxel value is determined to be stationary, the accumulated state data of that spatial voxel is decayed. For example, in this embodiment, when the voxel value is determined to be active, the accumulated state data of that spatial voxel is incremented by 1. When the voxel value is determined to be stationary, the accumulated state data of that spatial voxel is decremented by 1. This embodiment, by introducing an accumulation and decay mechanism, enables the accumulated state data to reflect the motion state of the object in the target three-dimensional space after statistical analysis of multiple consecutive frames. For example, if a spatial voxel remains empty for an extended period, the accumulated state data of the region represented by that voxel will decay to a lower value. Conversely, if a spatial voxel remains active, indicating the presence of an object within it, the accumulated state data will accumulate to a higher value. This allows for the differentiation between dynamic and static objects in space based on the accumulated state data of the spatial voxels. It's easy to understand that after statistical analysis using the accumulation and decay mechanism across multiple frames, the accumulated state data for fixed and moving objects in the target 3D space will differ. For instance, a stationary object in the target 3D space will accumulate a higher value after multiple frames of accumulation and decay, while a dynamic object will undergo accumulation and decay, resulting in a difference in its accumulated state data compared to a stationary object. It should be noted that in this embodiment, the range of the accumulated state data is [0, 100], meaning the minimum value after multiple decays is 0, and the maximum value after multiple accumulations is 100.

[0079] It should be noted that, in some embodiments of the present invention, the accumulation coefficient and attenuation coefficient of the accumulation and attenuation mechanisms can be customized according to downstream applications. For example, when motion data is dense and frequent in the detection scene, the attenuation coefficient and accumulation coefficient can be increased to highlight the characteristics of moving and stationary objects, effectively improving the ability of accumulated state data to indicate the motion state of objects.

[0080] In some embodiments of the present invention, after performing the step of identifying the dynamic object in the target three-dimensional space based on the voxel motion coefficients, the three-dimensional space dynamic recognition method provided in this embodiment further includes, but is not limited to:

[0081] A four-dimensional point cloud is constructed based on voxel motion coefficients and point cloud data.

[0082] The four-dimensional point cloud is input into a preset clustering algorithm for cluster analysis to obtain the three-dimensional data of the dynamic object. The three-dimensional data includes the length, width, and height data of the dynamic object.

[0083] In this specific embodiment, after identifying dynamic objects in the target 3D space based on voxel motion coefficients, the dynamic objects are filtered out from the target 3D space using a preset clustering algorithm. Specifically, this embodiment first combines voxel motion coefficients with point cloud data to construct a four-dimensional point cloud. This embodiment uses the voxel motion coefficients of each spatial voxel as the fourth dimension data and fuses it with the 3D point cloud data to obtain the four-dimensional point cloud data, i.e., the four-dimensional point cloud, to effectively improve the clustering effect. Further, this embodiment inputs the four-dimensional point cloud into the preset clustering algorithm for cluster analysis to obtain the 3D data of the dynamic objects. Specifically, the clustering algorithm can divide a set of physical or abstract objects into multiple classes composed of similar objects. Among them, the cluster generated by the clustering algorithm is a set of data objects that are similar to objects in the same cluster and different from objects in other clusters. In this embodiment, the constructed four-dimensional point cloud is used as the input of the preset clustering algorithm to perform cluster analysis on dynamic and static objects in the target 3D space. Based on the difference in voxel motion coefficients between dynamic and static objects, dynamic objects can be effectively filtered out from the target 3D space. Simultaneously, by combining the point cloud data corresponding to the spatial voxels—that is, the 3D data of each cloud point corresponding to the spatial voxels—the 3D data of the dynamic object can be calculated, such as its length, width, and height.

[0084] It is easy to understand that, in some embodiments of the present invention, after filtering out dynamic objects and obtaining corresponding three-dimensional data through a preset clustering algorithm, this embodiment can also combine some multi-target tracking algorithms to calculate the time-related data of each moving object in the detection of multiple consecutive frames, such as the movement speed and movement distance of the dynamic object.

[0085] In some embodiments of the present invention, the preset clustering algorithm includes a density-based clustering algorithm. Accordingly, the four-dimensional point cloud is input into the preset clustering algorithm for cluster analysis to obtain the three-dimensional data of the dynamic object, including but not limited to:

[0086] Three-dimensional data are obtained by performing cluster analysis on four-dimensional point clouds using density clustering algorithms.

[0087] In this specific embodiment, a density clustering algorithm is used to perform clustering analysis on the four-dimensional point cloud in the target three-dimensional space. Specifically, the density clustering algorithm defines a cluster as the largest set of density-connected points, which can divide regions with sufficiently high density into clusters and can perform clustering of various shapes in noisy spatial databases, effectively improving the accuracy and reliability of clustering analysis. In this embodiment, the four-dimensional point cloud is input into a density clustering algorithm, such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise), to perform clustering analysis on the point cloud in the target three-dimensional space. After analysis by the DBSCAN algorithm, the cloud point sets of each dynamic object are obtained. In this embodiment, the three-dimensional data of each dynamic object is calculated based on the three-dimensional information contained in each cloud point set.

[0088] An embodiment of the present invention also provides a three-dimensional spatial dynamic object recognition system, comprising:

[0089] The acquisition module is used to acquire point cloud data of the target's three-dimensional space.

[0090] The voxelization module is used to voxelize the target three-dimensional space to obtain several spatial voxels.

[0091] The calculation module is used to calculate the voxel value of each spatial voxel in the current point cloud frame based on the point cloud data. The voxel value is determined based on the point cloud data in the spatial voxel.

[0092] The update module is used to update the cumulative state data of each spatial voxel based on the voxel value.

[0093] The conversion module is used to convert accumulated state data into voxel motion coefficients using a preset motion indicator function. This preset motion indicator function includes a nonlinear function.

[0094] The recognition module is used to identify dynamic objects in the target three-dimensional space based on voxel motion coefficients.

[0095] Reference Figure 2 An embodiment of the present invention also provides a three-dimensional spatial dynamic object recognition system, comprising:

[0096] At least one processor 210.

[0097] At least one memory 220 is used to store at least one program.

[0098] When at least one program is executed by at least one processor 210, the at least one processor implements the three-dimensional spatial dynamic object recognition method as described in the above embodiments.

[0099] An embodiment of the present invention also provides a computer-readable storage medium storing computer-executable instructions that are executed by one or more control processors, for example, performing the steps described in the above embodiments.

[0100] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0101] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A method for recognizing dynamic objects in three-dimensional space, characterized in that, Includes the following steps: Acquire point cloud data of the target in three-dimensional space; The target three-dimensional space is voxelized to obtain several spatial voxels; The voxel value of each spatial voxel in the current point cloud frame is calculated based on the point cloud data; wherein the voxel value is determined based on the point cloud data in the spatial voxel; Update the cumulative state data of each spatial voxel based on the voxel value; The accumulated state data is converted into voxel motion coefficients using a preset motion indication function; wherein, the preset motion indication function includes a nonlinear function; Dynamic objects in the target three-dimensional space are identified based on the voxel motion coefficients; The step of calculating the voxel value of each spatial voxel in the current point cloud frame based on the point cloud data includes: The distribution data of the point cloud in each spatial voxel in the current point cloud frame is determined based on the point cloud data. Based on the distribution data, it is determined that a point cloud exists within the first spatial voxel, and the voxel value of the first spatial voxel is set to an active state. Based on the distribution data, it is determined that there is no point cloud within the second spatial voxel, and the voxel value of the second spatial voxel is set to a static state.

2. The three-dimensional dynamic object recognition method according to claim 1, characterized in that, The acquisition of point cloud data in the target three-dimensional space includes: A preset point cloud acquisition device is set in the target three-dimensional space; the preset point cloud acquisition device includes a lidar, a structured light scanning device, and a binocular stereo vision device. The point cloud data of the target three-dimensional space is acquired in real time by the preset point cloud acquisition device.

3. The three-dimensional dynamic object recognition method according to claim 1, characterized in that, The process of voxelizing the target three-dimensional space to obtain several spatial voxels includes: The voxel size data of the spatial voxel is determined according to preset voxel conditions; wherein, the preset voxel conditions include preset error tolerance, system computing power, and resolution of the point cloud data; The target three-dimensional space is voxelized based on the voxel size data to obtain the plurality of spatial voxels.

4. The three-dimensional dynamic object recognition method according to claim 1, characterized in that, The step of updating the cumulative state data of each spatial voxel based on the voxel value includes: The voxel value of the spatial voxel is determined to be the active state, and the cumulative data of the state of the spatial voxel is accumulated. Alternatively, the voxel value of the spatial voxel is determined to be the static state, and the accumulated state data of the spatial voxel is attenuated.

5. The three-dimensional dynamic object recognition method according to claim 1, characterized in that, After performing the step of identifying the dynamic object in the target three-dimensional space based on the voxel motion coefficients, the method further includes: A four-dimensional point cloud is constructed based on the voxel motion coefficients and the point cloud data; The four-dimensional point cloud is input into a preset clustering algorithm for cluster analysis to obtain the three-dimensional data of the dynamic object; wherein, the three-dimensional data includes the length data, width data and height data of the dynamic object.

6. The three-dimensional dynamic object recognition method according to claim 5, characterized in that, The preset clustering algorithm includes a density clustering algorithm; The step of inputting the four-dimensional point cloud into a preset clustering algorithm for cluster analysis to obtain the three-dimensional data of the dynamic object includes: The three-dimensional data is obtained by performing cluster analysis on the four-dimensional point cloud using the density clustering algorithm.

7. A three-dimensional spatial dynamic object recognition system, characterized in that, include: The acquisition module is used to acquire point cloud data of the target's three-dimensional space; A voxelization module is used to voxelize the target three-dimensional space to obtain several spatial voxels; The calculation module is used to calculate the voxel value of each spatial voxel in the current point cloud frame based on the point cloud data; wherein the voxel value is determined based on the point cloud data in the spatial voxel; The update module is used to update the cumulative state data of each spatial voxel based on the voxel value; A conversion module is used to convert the accumulated state data into voxel motion coefficients using a preset motion indication function; wherein the preset motion indication function includes a nonlinear function. The identification module is used to identify dynamic objects in the target three-dimensional space based on the voxel motion coefficients; The step of calculating the voxel value of each spatial voxel in the current point cloud frame based on the point cloud data includes: The distribution data of the point cloud in each spatial voxel in the current point cloud frame is determined based on the point cloud data. Based on the distribution data, it is determined that a point cloud exists within the first spatial voxel, and the voxel value of the first spatial voxel is set to an active state. Based on the distribution data, it is determined that there is no point cloud within the second spatial voxel, and the voxel value of the second spatial voxel is set to a static state.

8. A three-dimensional spatial dynamic object recognition system, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the three-dimensional spatial dynamic object recognition method as described in any one of claims 1 to 6.

9. A computer storage medium storing a processor-executable program, characterized in that, The program executable by the processor is used, when executed by the processor, to implement the three-dimensional spatial dynamic object recognition method as described in any one of claims 1 to 6.

Images