Mobile target identification method and device, computer device, medium and program product

Abstract

The present application relates to the technical field of computer, disclose mobile target identification method, device, computer equipment, medium and program product, the method provided by the present application, based on the preset sliding time window obtains a plurality of point cloud time series, utilize a plurality of point cloud time series to carry out subsequent mobile target identification, it does not need to be based on a complete time period to carry out subsequent calculation to target object scanning result, effectively improve the real-time performance of prediction, and convert the perspective of the second point cloud in each time sequence in the time window to the first point cloud, output each converted point cloud time series to the calculation model constructed in advance, calculate the confidence score corresponding to each point cloud time series respectively, and fuse the confidence score corresponding to each point cloud time series respectively by using the preset filter, based on the fusion result, mobile target identification is carried out, the estimation effect of dynamic object is effectively improved, and the accuracy of mobile target identification result is improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically to methods, apparatus, computer equipment, media, and program products for identifying moving targets. Background Technology

[0002] Robots need to recognize the navigability of their local environment during movement, which is achieved through LiDAR modeling of various elements in the 3D environment. Moving target segmentation in 3D LiDAR data does not require complex semantic class concepts and extensive labeling. Instead, the goal of these works is to predict whether the local point cloud structure moves or remains static throughout space and time, i.e., to determine which points in a scan sequence are moving and which are not.

[0003] In related technologies, offline estimation methods are generally used for moving target estimation. Offline estimation methods rely on acquiring observation results from all sequences of lidar data, which consumes one-time computing resources. Furthermore, these methods require back-projection or clustering based on complex calculations to retrieve the prediction for each point in point estimation, resulting in poor generalization ability to the location environment. Summary of the Invention

[0004] In view of this, the present invention provides a moving target identification method, apparatus, computer equipment, medium and program product to solve the problem of computational complexity in moving target estimation based on offline estimation methods in related technologies.

[0005] In a first aspect, the present invention provides a moving target identification method, the method comprising: acquiring multiple point cloud time series based on a preset sliding time window, each point cloud time series including a first point cloud and at least one second point cloud, wherein the time of the first point cloud is earlier than the time of the second point cloud; converting the second point cloud in each point cloud time series to the perspective of the first point cloud to obtain corresponding converted point cloud time series data; inputting each converted point cloud time series data into a pre-constructed computational model to calculate confidence scores corresponding to different point clouds in each point cloud time series; fusing the confidence scores corresponding to different point clouds in each point cloud time series using a preset filter to obtain a motion evaluation value for the corresponding point cloud time series; determining at least one target point cloud time series from the multiple point cloud time series based on the motion evaluation values ​​corresponding to different point cloud time series, wherein the motion evaluation value of the target point cloud time series is greater than a preset threshold; and determining a point cloud set of a moving target based on each target point cloud time series.

[0006] The moving target recognition method provided by this invention acquires multiple point cloud time series through a preset sliding time window, transforms the second point cloud in each point cloud time series to the perspective of the first point cloud, outputs the transformed point cloud time series data to a pre-constructed computing model, calculates the confidence scores corresponding to different point clouds in each point cloud time series, fuses the estimation results corresponding to different point clouds in each point cloud time series using a preset filter, obtains the movement evaluation value of the corresponding point cloud time series, determines at least one target point cloud time series from the multiple point cloud time series based on the movement evaluation values ​​corresponding to different point cloud time series, the movement evaluation value of the target point cloud time series is greater than a preset threshold, and determines the point cloud set of the moving target based on each target point cloud time series. The method provided by this invention acquires multiple point cloud time series based on a preset sliding time window, and uses the converted multiple point cloud time series for subsequent moving target recognition. It does not require subsequent calculations based on the target object scanning results of each frame within the time period, effectively improving the real-time performance of the prediction. Based on the moving confidence score prediction model, it predicts the confidence scores corresponding to different point clouds in each point cloud time series, and uses a preset filter to fuse the confidence scores corresponding to different point clouds in each point cloud time series. Based on the fusion result, it determines the point cloud set of the moving target, effectively improving the perception effect and the accuracy of the moving target recognition results.

[0007] In one optional implementation, the step of acquiring multiple point cloud time series based on a preset sliding time window includes: acquiring an initial point cloud time series collected by a laser detection and ranging system; preprocessing the initial point cloud time series to obtain a preprocessed target point cloud time series; and determining multiple point cloud time series based on the preset time window and the target point cloud time series.

[0008] In one optional implementation, the step of transforming the second point cloud in each point cloud time series to the perspective of the first point cloud to obtain the corresponding transformed point cloud time series data includes: determining the transition parameters between the first point cloud and the second point cloud in the corresponding point cloud time series based on the first point cloud and the second point cloud in each point cloud time series; and determining the corresponding transformed point cloud time series based on the transition parameters between the first point cloud and the second point cloud in each point cloud time series.

[0009] In one optional implementation, the step of determining the point cloud set of a moving target based on the time series of each target point cloud includes: determining multiple target points in each target point cloud time series, wherein the target points are points in each second point cloud in the corresponding target point cloud time series that are increased or decreased relative to the first point cloud; and determining the point cloud set of the moving target based on the multiple target points in each target point cloud time series.

[0010] In one optional implementation, the step of preprocessing the initial point cloud time series to obtain the preprocessed target point cloud time series includes: using a preset algorithm to remove ground point clouds, noise and outliers from the initial point cloud time series to obtain the processed first point cloud time series; and determining the preprocessed target point cloud time series based on the first point cloud time series.

[0011] In one optional implementation, the step of determining the preprocessed target point cloud time series based on the first point cloud time series includes: spatially partitioning the first point cloud time series using the voxel method to obtain a processing result; determining the target array matrix based on the processing result; performing format transformation on the target array matrix using a preset method to obtain a second point cloud time series; and registering the point cloud data at each time point in the second point cloud time series to obtain the target point cloud time series.

[0012] Secondly, the present invention provides a moving target identification device, comprising: an acquisition module for acquiring multiple point cloud time series based on a preset time window, each point cloud time series including a first point cloud and at least one second point cloud, wherein the time of the first point cloud is earlier than the time of the second point cloud; a conversion module for converting the second point cloud in each point cloud time series to the perspective of the first point cloud, thereby obtaining corresponding converted point cloud time series data; a first determination module for inputting each converted point cloud time series data into a pre-constructed calculation model, thereby calculating confidence scores corresponding to different point clouds in each point cloud time series; a fusion module for fusing the confidence scores corresponding to different point clouds in each point cloud time series using a preset filter, thereby obtaining a motion evaluation value for the corresponding point cloud time series; a second determination module for determining at least one target point cloud time series from the multiple point cloud time series based on the motion evaluation values ​​corresponding to different point cloud time series, wherein the motion evaluation value of the target point cloud time series is greater than a preset threshold; and a third determination module for determining a point cloud set of a moving target based on each target point cloud time series.

[0013] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the moving target recognition method of the first aspect or any corresponding embodiment described above.

[0014] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the moving target recognition method of the first aspect or any corresponding embodiment described above.

[0015] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the moving target recognition method of the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a moving target identification method according to an embodiment of the present invention;

[0018] Figure 2 This is a flowchart illustrating another moving target identification method according to an embodiment of the present invention;

[0019] Figure 3 This is a flowchart illustrating another moving target identification method according to an embodiment of the present invention;

[0020] Figure 4 This is a structural block diagram of a moving target identification device according to an embodiment of the present invention;

[0021] Figure 5 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] In related technologies, offline estimation methods are generally used for moving target estimation. Offline estimation methods rely on acquiring observation results from all sequences of lidar data, which consumes one-time computing resources. Furthermore, these methods require back-projection or clustering based on complex calculations to retrieve the prediction for each point in point estimation, resulting in poor generalization ability to the location environment.

[0024] In view of this, the moving target recognition method provided in this application embodiment can be applied to a server to realize the judgment of moving targets. The method provided by the present invention acquires multiple point cloud time series based on a preset sliding time window, and uses the converted multiple point cloud time series for subsequent moving target recognition. It does not require subsequent calculation based on the target object scanning results of each frame within the time period, which effectively improves the real-time performance of prediction. Based on the moving confidence score prediction model, it predicts the confidence scores corresponding to different point clouds in each point cloud time series, and uses a preset filter to fuse the confidence scores corresponding to different point clouds in each point cloud time series. Based on the fusion result, it determines the point cloud set of the moving target, which effectively improves the perception effect and the accuracy of the moving target recognition result.

[0025] According to an embodiment of the present invention, a method for identifying moving targets is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0026] This embodiment provides a moving target identification method, which can be used in the aforementioned server. Figure 1 This is a flowchart of a moving target recognition method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0027] Step S101: Obtain multiple point cloud time series based on a preset sliding time window.

[0028] For example, each point cloud time series includes a first point cloud and at least one second point cloud, where the time of the first point cloud is later than the time of the second point cloud. The preset sliding time window can be any suitable length, and it employs a back-off window strategy. In this embodiment, based on the back-off window strategy, the length of the preset sliding time window can be 5 time nodes, and the first point cloud is the point cloud corresponding to the earliest time in the point cloud time series. The advantage of the back-off window strategy is that it allows for the re-estimation of moving targets in the scene by using new observation data from the input sequence.

[0029] Step S102: Transform the second point cloud in each point cloud time series to the perspective of the first point cloud to obtain the corresponding transformed point cloud time series data.

[0030] For example, in this embodiment of the application, the second point cloud in each point cloud time series is transformed to the perspective of the first point cloud, thereby unifying the perspective of point clouds at different time nodes in each point cloud time series, which facilitates the subsequent real-time judgment of target objects.

[0031] Step S103: Input the transformed point cloud time series data into the pre-built calculation model, and calculate the confidence scores corresponding to different point clouds in each point cloud time series.

[0032] For example, a pre-built computational model can be obtained by training a neural network model. This computational model is used to characterize the correlation between the transformed point cloud time series and the confidence score. In this embodiment, an improved sparse 4D convolutional neural network is used, which is a modification of the MinkUNet14 framework. The number of feature channels in the network is reduced, and a high-dimensional network structure is constructed using generalized sparse convolution with custom stride and kernel shape. Specifically, a neural network with four-dimensional convolution is used, treating the time dimension as an additional spatial dimension. Efficient sparse convolution extracts the spatiotemporal features of the sparse 4D grid and predicts the moving target confidence score for each point. Compared with dense convolution, sparse convolution is computationally faster. During 4D convolution, the convolution kernel has a size of 3 on the time scale. During the stride, the 4D kernel is applied to consecutive 3D cubes to extract feature information of the entire time series from the previous layer. Sparse convolution acts on sparse tensors and defines a kernel mapping relationship, which specifies the kernel weights connecting the input and output coordinates. The convolutional neural network used in this embodiment is a residual bottleneck architecture. It uses strided sparse convolution to downsample the feature map, uses cross-sparse transpose convolution to upsample, and uses skip connections to help obtain feature details and achieve fine-grained prediction. The last layer of the neural network is a 4D sparse convolution with a softmax function to predict the confidence score of a moving target, with the result ranging from 0 to 1.

[0033] Step S104: Use a preset filter to fuse the confidence scores corresponding to different point clouds in each point cloud time series to obtain the motion evaluation value of the corresponding point cloud time series.

[0034] For example, in this embodiment of the application, the preset filter can be a binary Bayesian filter. The binary Bayesian filter is used to fuse the confidence scores corresponding to different point clouds in each point cloud time series to obtain multiple estimation results. Specifically, for the transformed point cloud time series... The point cloud data at each time point follows Bayes' theorem, and its joint probability density can be expressed as shown in equation (1):

[0035]

[0036] in, Represents state variables; Represents the transformed point cloud time series The joint probability density function of all points.

[0037] Furthermore, after deriving the standard recursive binary Bayesian filter, the motion evaluation value of the target object can be determined by the following equation (2):

[0038]

[0039] Where Γ is the input sequence z t The set of time steps for observation point p, where observation point p represents a point in the point cloud, the point currently being evaluated; l(m i |z 0:t-1 ) is a recursive term that includes all predictions from time i to time t-1; l(m i |z 0:t The ) represents the logarithmic ratio of the probability of moving at time t. (The last part...) An innovative metric is provided for prediction, where p0 represents the prior probability. For moving target segmentation, the prior probability determines the degree to which the predicted moving points in a single scan influence the final prediction. At time t, the sparse convolutional neural network outputs the confidence score of the moving target. The posterior probability is represented by ξ t,i =p(m i =1|z t The current input sequence is z. t M represents the point cloud S j Given the set of points, the above logarithmic ratio formula is shown in equation (3) below:

[0040]

[0041] Step S105: Based on the motion evaluation values ​​corresponding to different point cloud time series, at least one target point cloud time series is determined from the plurality of point cloud time series, wherein the motion evaluation value of the target point cloud time series is greater than a preset threshold. Exemplarily, in this embodiment, the moving object is determined based on the magnitude of the motion evaluation value.

[0042] The moving target recognition method provided in this embodiment acquires multiple point cloud time series based on a preset sliding time window, and uses the converted multiple point cloud time series for subsequent moving target recognition. It does not require subsequent calculations based on the target object scanning results of each frame within the time period, effectively improving the real-time performance of the prediction. Based on the moving confidence score prediction model, it predicts the confidence scores corresponding to different point clouds in each point cloud time series, and uses a preset filter to fuse the confidence scores corresponding to different point clouds in each point cloud time series. Based on the fusion result, it determines the point cloud set of the moving target, effectively improving the perception effect and the accuracy of the moving target recognition results.

[0043] This embodiment provides a moving target identification method, which can be used in the aforementioned server. Figure 2 This is a flowchart of a moving target recognition method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps:

[0044] Step S201: Obtain multiple point cloud time series based on a preset sliding time window. Each point cloud time series includes a first point cloud and at least one second point cloud, where the time of the first point cloud is earlier than the time of the second point cloud. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0045] Specifically, step S201 includes:

[0046] Step S2011: Obtain the initial point cloud time series collected by the laser detection and ranging system.

[0047] For example, in this embodiment of the application, the lidar device is electrically connected to the industrial control computer through a dedicated connector, the corresponding driver is installed, the network is configured on the industrial control computer, and the IP address, subnet mask, and gateway address of the lidar are entered sequentially in the configuration interface; then, the status, motor, field of view and other parameters can be set by entering the lidar IP address in the browser, and a temporary route can be set by adding the lidar address through the command line; after that, a project file is created, and the point cloud data can be displayed and saved in real time by entering the corresponding commands.

[0048] Step S2012: Preprocess the initial point cloud time series to obtain the preprocessed target point cloud time series.

[0049] In some optional implementations, step S2012 above includes:

[0050] Step a1: Use a preset algorithm to remove ground point clouds, noise and outliers from the initial point cloud time series to obtain the processed first point cloud time series.

[0051] For example, in this embodiment, an angle threshold algorithm is used to remove ground point clouds, that is, to distinguish ground points by utilizing the angle difference formed by the laser scan line between the ground and the target object. Multi-line lidar sensors form larger angles between objects of the same type and smaller angles between adjacent objects during horizontal scanning. This characteristic is used to set an angle threshold to remove ground point clouds. Noise is usually unrelated to the target representation. Convolution is used to process noisy point cloud data, preserving the feature information of the point cloud model to the maximum extent while meeting real-time requirements. Projection filtering technology is chosen to adjust the position of each point in the point cloud by projecting it into a low-dimensional space. Points with abnormal intensity and number of bounces are identified by judging critical values, thereby achieving noise removal. For outliers, this invention uses statistical methods for removal. For a point in the point cloud, several nearest neighbor points are selected, and statistical parameters such as the standard deviation ratio of the point cloud are calculated. If a point deviates from the average value by n times, it is considered an outlier and needs to be deleted.

[0052] Step a2: Determine the preprocessed target point cloud time series based on the first point cloud time series.

[0053] In some alternative implementations, step a2 above includes:

[0054] Step a21: Spatial partitioning of the first point cloud time series is performed using the voxel method to obtain the processing result. Exemplarily, in this embodiment, the voxel method is used to spatially partition the point cloud. By calculating the number of adjacent points, an octree 3D mesh model is used to check for any blank points around the point cloud. If blank points are found, the voxel surface points are retained and converted to Cartesian coordinates, then saved in .ply format.

[0055] Step a22: Determine the target array matrix based on the processing result. For example, the target array matrix is ​​determined based on the coordinates of points in the first point cloud time series after format partitioning.

[0056] Step a23: Use a preset method to perform format transformation on the target array matrix to obtain the second point cloud time series.

[0057] For example, in the embodiments of this application, the preset method may include, but is not limited to, matrix calculation method.

[0058] Step a24: Register the point cloud data at each time point in the second point cloud time series to obtain the target point cloud time series.

[0059] For example, in the embodiments of this application, key points are extracted from different point cloud data obtained at the same time according to the same benchmark, feature descriptors of these key points are calculated, and the corresponding point pairs are initially estimated by combining the position of the descriptors in the dataset and the similarity between features and positions; the rigid body transformation relationship is estimated based on the point correspondence relationship, and finally a suitable coordinate transformation (rotation matrix R and translation vector T) is determined, and the point cloud sets obtained from various perspectives are merged into a unified coordinate system.

[0060] Step S2013: Determine multiple point cloud time series based on the preset time window and the target point cloud time series.

[0061] Step S202: Transform the second point cloud in each point cloud time series to the perspective of the first point cloud to obtain the corresponding transformed point cloud time series.

[0062] Specifically, step S202 includes:

[0063] Step S2021: Based on the first point cloud and the second point cloud in each point cloud time series, determine the transition parameters between the first point cloud and the second point cloud in the corresponding point cloud time series.

[0064] For example, point cloud representation, given a set of point cloud sequences This represents an N-line lidar scan. First, the iterative least squares method is used to solve the sequence S1, ..., S2 in the point cloud set S0 of the current lidar scan results. N 1. Perform local alignment adjustments and scan result set. Where p i =[x i ,y i ,z i ,1] Τ Let R denote the set of real numbers. Assume S j 1 and S j The attitude between them is the transition parameter Transfer parameters Together they form the transition matrix Define the input representation of the j-th scan result transformed into the current viewpoint as: in k represents a number between 0 and j-1.

[0065] Step S2022: Based on the transition parameters between the first and second point clouds in each point cloud time series, determine the corresponding transformed point cloud time series.

[0066] For example, the converted point cloud time series is a 4D point cloud sequence. In this embodiment, coordinate transformation and time are used as one dimension to aggregate the scanning results to form a 4D point cloud. This embodiment quantizes the 4D point cloud into a sparse voxel grid with a fixed resolution in time t and space s, represented by a sparse tensor. The point cloud input only retains the voxels occupied by points. The convolution, pooling, and other operations of the sparse tensor are based on the Minkowski engine, that is, the 4D point cloud is divided into sparse voxel space by combining and summing the sparse points in space, and sparse voxels are represented by sparse tensors. Only the spatial voxel features with objects are expressed to improve the efficiency of the algorithm, allowing the algorithm to be used in large spaces. The advantage of this representation is that when the current network is deployed in a new environment, it is not necessary to consider the two-dimensional coordinate distribution or standardize the data.

[0067] Step S203 involves inputting the transformed point cloud time series data into a pre-built computational model to calculate the confidence scores corresponding to different point clouds in each time series. For details, please refer to [link to relevant documentation]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0068] Step S204: A preset filter is used to fuse the confidence scores corresponding to different point clouds in each point cloud time series to obtain the motion evaluation value of the corresponding point cloud time series. For details, please refer to [link to relevant documentation]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.

[0069] Step S205: Based on the motion evaluation values ​​corresponding to different point cloud time series, determine at least one target point cloud time series from the plurality of point cloud time series, wherein the motion evaluation value of the target point cloud time series is greater than a preset threshold. For details, please refer to [link to relevant documentation]. Figure 1 Step S105 of the illustrated embodiment will not be described again here.

[0070] Step S206: Determine the point cloud set of the moving target based on the time series of each target point cloud.

[0071] This embodiment provides a moving target identification method, which can be used in the aforementioned server. Figure 3 This is a flowchart of a moving target recognition method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps:

[0072] Step S301: Obtain multiple point cloud time series based on a preset sliding time window. Each point cloud time series includes a first point cloud and at least one second point cloud, where the time of the first point cloud is earlier than the time of the second point cloud. For details, please refer to [link to relevant documentation]. Figure 1 Step S201 of the illustrated embodiment will not be described again here.

[0073] Step S302: Transform the second point cloud in each point cloud time series to the perspective of the first point cloud to obtain the corresponding transformed point cloud time series data. For details, please refer to [link to relevant documentation]. Figure 2 Step S202 of the illustrated embodiment will not be described again here.

[0074] Step S303: Input the transformed point cloud time series data into the pre-built computational model, and calculate the confidence scores corresponding to different point clouds in each point cloud time series. For details, please refer to... Figure 2 Step S203 of the illustrated embodiment will not be described again here.

[0075] Step S304: The confidence scores corresponding to different point clouds in each point cloud time series are fused using a preset filter to obtain the motion evaluation value of the corresponding point cloud time series. For details, please refer to [link to relevant documentation]. Figure 2 Step S204 of the illustrated embodiment will not be described again here.

[0076] Step S305: Based on the motion evaluation values ​​corresponding to different point cloud time series, determine at least one target point cloud time series from the plurality of point cloud time series, wherein the motion evaluation value of the target point cloud time series is greater than a preset threshold. For details, please refer to [link to relevant documentation]. Figure 2 Step S204 of the illustrated embodiment will not be described again here.

[0077] Step S306: Determine the point cloud set of the moving target based on the time series of each target point cloud.

[0078] Specifically, step S306 includes:

[0079] Step S3061: Determine multiple target points in each target point cloud time series. The target points are the points in the corresponding target point cloud time series where the second point cloud increases or decreases relative to the first point cloud.

[0080] Step S3062: Based on multiple target points in the time series of each target point cloud, the point cloud set of the moving target is formed.

[0081] For example, in the embodiments of this application, the preset threshold may include, but is not limited to, 0.5. If the confidence score is greater than 0.5, then the points in the target point cloud time series that are increased or decreased relative to the first point cloud are determined as target points. In order to quantitatively evaluate the effectiveness of the method, this invention uses the standard intersection-over-union ratio (IoU), that is, the overlap rate between the target window generated by the model and the original labeled window, to evaluate the detection accuracy, i.e.:

[0082]

[0083] In this dataset, TP, FP, and FN represent true positive, false positive, and false negative sample points, respectively. The IoU was obtained from datasets such as SemanticKITTI. MOS =65.2% of the results.

[0084] This embodiment also provides a moving target identification device for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0085] This embodiment provides a moving target identification device, such as... Figure 4 As shown, it includes:

[0086] The acquisition module 401 is used to acquire multiple point cloud time series based on a preset time window. Each point cloud time series includes a first point cloud and at least one second point cloud, and the time of the first point cloud is earlier than the time of the second point cloud.

[0087] The conversion module 402 is used to convert the second point cloud in each point cloud time series to the perspective of the first point cloud, so as to obtain the corresponding converted point cloud time series data.

[0088] The first determining module 403 is used to input the transformed point cloud time series data into the pre-built calculation model and calculate the confidence scores corresponding to different point clouds in each point cloud time series.

[0089] The fusion module 404 is used to fuse the confidence scores corresponding to different point clouds in each point cloud time series using a preset filter to obtain the motion evaluation value of the corresponding point cloud time series.

[0090] The second determining module 405 determines at least one target point cloud time series from the plurality of point cloud time series using the motion evaluation values ​​corresponding to different point cloud time series, wherein the motion evaluation value of the target point cloud time series is greater than a preset threshold.

[0091] The third determining module 406 is used to determine the point cloud set of the moving target based on the time series of each target point cloud.

[0092] In some optional implementations, the acquisition module 401 includes:

[0093] The acquisition submodule is used to acquire the initial point cloud time series collected by the laser detection and ranging system;

[0094] The preprocessing submodule is used to preprocess the initial point cloud time series to obtain the preprocessed target point cloud time series;

[0095] The first determination submodule is used to determine multiple point cloud time series based on a preset time window and the target point cloud time series.

[0096] In some alternative implementations, the conversion module 402 includes:

[0097] The second determining submodule is used to determine the transition parameters between the first point cloud and the second point cloud in the corresponding point cloud time series based on the first point cloud and the second point cloud in each point cloud time series.

[0098] The third determination submodule is used to determine the corresponding transformed point cloud time series based on the transition parameters between the first and second point clouds in each point cloud time series.

[0099] In some alternative implementations, the second determining module 406 includes:

[0100] The fourth determination submodule is used to determine multiple target points in each target point cloud time series. The target points are the points in the corresponding target point cloud time series that are increased or decreased relative to the first point cloud in each second point cloud.

[0101] The fifth determination submodule is used to move the point cloud set of the target based on multiple target points in the time series of each target point cloud.

[0102] In some alternative implementations, the preprocessing submodule includes:

[0103] The elimination unit is used to remove ground point clouds, noise and outliers from the initial point cloud time series using a preset algorithm to obtain the processed first point cloud time series.

[0104] The first determining unit is used to determine the preprocessed target point cloud time series based on the first point cloud time series.

[0105] In some optional implementations, the first determining unit includes:

[0106] The spatial sub-units are used to perform spatial partitioning of the first point cloud time series using the voxel method to obtain the processing results.

[0107] Determine the sub-units, which are used to determine the target array matrix based on the processing results;

[0108] The format transformation subunit performs format transformation on the target array matrix using a preset method to obtain the second point cloud time series;

[0109] The registration subunit is used to register the point cloud data at each time point in the second point cloud time series to obtain the target point cloud time series.

[0110] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0111] The moving target identification device in this embodiment is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0112] This invention also provides a computer device having the above-described features. Figure 4 The moving target identification device shown.

[0113] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 5 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 5 Take a processor 10 as an example.

[0114] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0115] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0116] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0117] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0118] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0119] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0120] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0121] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for identifying moving targets, characterized in that, The method includes: Multiple point cloud time series are obtained based on a preset sliding time window. Each point cloud time series includes a first point cloud and at least one second point cloud, with the time of the first point cloud being earlier than the time of the second point cloud. The second point cloud in each point cloud time series is transformed to the perspective of the first point cloud to obtain the corresponding transformed point cloud time series data. The transformed point cloud time series data are input into a pre-built computational model, and the confidence scores corresponding to different point clouds in each point cloud time series are calculated. The confidence scores of different point clouds in each point cloud time series are fused using a preset filter to obtain the motion evaluation value of the corresponding point cloud time series. Based on the motion evaluation values ​​corresponding to different point cloud time series, at least one target point cloud time series is determined from the plurality of point cloud time series, wherein the motion evaluation value of the target point cloud time series is greater than a preset threshold. The point cloud set of the moving target is determined based on the time series of each target point cloud; The steps for determining the point cloud set of a moving target based on the time series of each target point cloud include: Multiple target points are identified in each target point cloud time series, wherein the target points are the points in the corresponding target point cloud time series that are increased or decreased relative to the first point cloud; Based on multiple target points in the time series of each target point cloud, the point cloud set of the moving target is determined.

2. The method according to claim 1, characterized in that, The steps for obtaining multiple point cloud time series based on a preset sliding time window include: Obtain the initial point cloud time series acquired by the laser detection and ranging system; The initial point cloud time series is preprocessed to obtain the preprocessed target point cloud time series; The multiple point cloud time series are determined based on a preset time window and the target point cloud time series.

3. The method according to claim 1, characterized in that, The steps for transforming the second point cloud in each point cloud time series to the perspective of the first point cloud to obtain the corresponding transformed point cloud time series data include: Based on the first and second point clouds in each point cloud time series, determine the transition parameters between the first and second point clouds in the corresponding point cloud time series. Based on the transition parameters between the first and second point clouds in each point cloud time series, the corresponding transformed point cloud time series is determined.

4. The method according to claim 2, characterized in that, The step of preprocessing the initial point cloud time series to obtain the preprocessed target point cloud time series includes: The ground point cloud, noise, and outliers of the initial point cloud time series are removed using a preset algorithm to obtain the processed first point cloud time series; The preprocessed target point cloud time series is determined based on the first point cloud time series.

5. The method according to claim 4, characterized in that, The steps for determining the preprocessed target point cloud time series based on the first point cloud time series include: The first point cloud time series was spatially partitioned using the voxel method to obtain the processing result; The target array matrix is ​​determined based on the processing results; The target array matrix is ​​format-transformed using a preset method to obtain the second point cloud time series; The point cloud data at each time point in the second point cloud time series are registered to obtain the target point cloud time series.

6. A moving target identification device, characterized in that, The device includes: The acquisition module is used to acquire multiple point cloud time series based on a preset time window. Each point cloud time series includes a first point cloud and at least one second point cloud, and the time of the first point cloud is earlier than the time of the second point cloud. The conversion module is used to convert the second point cloud in each point cloud time series to the perspective of the first point cloud, so as to obtain the corresponding converted point cloud time series data. The first determining module is used to input the transformed point cloud time series data into the pre-built calculation model and calculate the confidence scores corresponding to different point clouds in each point cloud time series. The fusion module is used to fuse the confidence scores of different point clouds in each point cloud time series using a preset filter to obtain the motion evaluation value of the corresponding point cloud time series. The second determining module is used to determine at least one target point cloud time series from the plurality of point cloud time series based on the motion evaluation values ​​corresponding to different point cloud time series, wherein the motion evaluation value of the target point cloud time series is greater than a preset threshold. The third determining module is used to determine the point cloud set of the moving target based on the time series of each target point cloud; The point cloud set for determining the moving target based on the time series of each target point cloud includes: Multiple target points are identified in each target point cloud time series, wherein the target points are the points in the corresponding target point cloud time series that are increased or decreased relative to the first point cloud; Based on multiple target points in the time series of each target point cloud, the point cloud set of the moving target is determined.

7. A computer device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the moving target identification method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the moving target identification method according to any one of claims 1 to 5.

9. A computer program product, characterized in that, It includes computer instructions for causing a computer to perform the moving target identification method according to any one of claims 1 to 5.

Images