A method for point cloud target tracking and recognition and a lidar

Abstract

The present invention discloses a method for point cloud target tracking and recognition and a lidar, belonging to the technical field of data tracking and recognition, which reduces the data processing amount of point cloud target tracking and recognition and improves the efficiency of point cloud target tracking and recognition. The method includes: determining the next-frame point cloud target tracking and recognition parameters according to the current-frame point cloud target tracking and recognition result, and determining the next-frame point cloud processing increment range according to the next-frame point cloud target tracking and recognition parameters; updating the point cloud processing range to be tracked and recognized according to the current-frame point cloud processing range and the next-frame point cloud processing increment range; tracking and recognizing the point cloud target in the point cloud processing range to be tracked and recognized to obtain the point cloud target tracking and recognition result.

Description

Technical Field

[0001] The present invention relates to the technical field of data tracking and recognition, and in particular to a point cloud target tracking and recognition method and a lidar. Background Art

[0002] In the prior art, the main devices used for object detection include cameras, lidars, millimeter wave radars, etc. However, there are generally problems such as inability to work all day long, excessive acquisition of redundant information, large amount of data, and insufficient automation, resulting in certain missed detections and false alarms in the detection of objects in key areas. In order to solve the problems existing in the current detection means, this patent designs an improved object detection system to strengthen the recognition and detection of foreign objects, uses a lidar that works all day long as the main sensor, improves the stability of sensor monitoring and detection, and realizes all-day object detection.

[0003] Different from the previous use of lidar for object recognition, for continuously moving objects, the point cloud obtained by using lidar has the following characteristics: ① large amount of data, and the computational complexity of the lidar for processing a single frame of point cloud is very high. ② The difference between adjacent point cloud frames is small, and there are a large number of background point clouds whose positions have not changed. ③ The point cloud of the object to be detected usually moves only in a certain direction, and the position coordinate values of some point clouds between adjacent point cloud frames do not change. ④ The point cloud data is completely discrete, and the presence or absence of some point clouds does not affect other point clouds. ⑤ The point cloud data shows the characteristics of being dense near and sparse far away, with the point cloud obtained in the distance being sparse and the point cloud being denser near. In view of the problem of time-consuming processing of the current large amount of point cloud data and the characteristics of the point cloud of continuously moving objects, the object detection method is improved to reduce the computational burden of the lidar.

[0004] Existing lidar object detection methods are mainly divided into two categories. One is the detection method based on clustering, and the other is the detection method based on training. The main processes of the clustering-based method are ground point removal, point cloud clustering, point cloud segmentation, and object detection by point cloud size matching. The object recognition process based on training regards each frame as an independent input point cloud and relies on the feature extraction ability of the deep neural network to achieve object detection. In the construction of the deep neural network for point cloud object detection, there are generally two construction ideas. The first is the two-stage method represented by PointRCNN, that is, the entire deep neural network is regarded as a network composed of two stages: object localization and object classification. The training stage is also carried out in two stages. First, the training of object localization is carried out, and then the object target classification is carried out on the point cloud at the localization position obtained from the first-stage training. The second is the single-stage training process represented by 3DSSD, which directly identifies the point position and classification with a neural network to directly realize the training of object localization and classification. In the real-time positioning and tracking of actual moving objects, the input point cloud is generally preprocessed, mainly including the removal of point cloud outliers, the surface smoothing filtering of the point cloud, and appropriate cropping. Then, object recognition is realized on the trained point cloud object localization detection network.

[0005] However, whether it is the preprocessing of point cloud data or the tracking and detection of objects, since the detected point cloud objects usually only occupy a very small part of all the input point cloud data and the amount of irrelevant data is huge, the lidar processes a lot of featureless data. For example, when applying a multi-line lidar to the detection of foreign object intrusion in a railway scene, take the foreign objects that may appear in the scene as an example. In the railway monitoring range scanned by the multi-line lidar, the range covered by the lidar field of view scanning is a point cloud space with a length of about 100 meters, a width of 20 meters, and a height of 10 meters. However, the size of the pedestrian point cloud is only about 0.5 cubic meters in volume, and the proportion of the occupied space is less than 1%. The amount of environmental irrelevant point cloud data is nearly a hundred times that of the target object point cloud. However, all the input point clouds are processed equally, and the irrelevant feature points are processed multiple times, increasing the amount of computation and the computational burden on the lidar data processing module, which will slow down the detection speed of the lidar. For the detection of other foreign objects, the point cloud redundancy of small-sized irrelevant objects will be even greater, and these data redundancies will greatly increase the computational burden of the lidar. Summary of the Invention

[0006] In view of the above analysis, the embodiments of the present invention aim to provide a point cloud target tracking and recognition method and a lidar, which reduce the data processing amount of point cloud target tracking and recognition and improve the efficiency of point cloud target tracking and recognition.

[0007] The present invention discloses a point cloud target tracking and recognition method, including:

[0008] Determine the next-frame point cloud target tracking and recognition parameters based on the current-frame point cloud target tracking and recognition result, and determine the next-frame point cloud processing increment range according to the next-frame point cloud target tracking and recognition parameters;

[0009] Update the point cloud processing range to be tracked and recognized according to the current-frame point cloud processing range and the next-frame point cloud processing increment range;

[0010] Track and recognize the point cloud target in the point cloud processing range to be tracked and recognized, and obtain the point cloud target tracking and recognition result.

[0011] Further, the determining the next-frame point cloud target tracking and recognition parameters according to the current-frame point cloud target tracking and recognition result includes:

[0012] Determine the distance between the current-frame point cloud target and the lidar, and use the distance, as well as the maximum and minimum values of the current-frame point cloud target in each coordinate component, as the next-frame point cloud target tracking and recognition parameters;

[0013] And / or, determine the current-frame point cloud data volume corresponding to the current-frame point cloud target, and use the ratio between the current-frame point cloud data volume and the background point cloud data volume as the next-frame point cloud target tracking and recognition parameter.

[0014] Further, the updating the point cloud processing range to be tracked and recognized according to the current-frame point cloud processing range and the next-frame point cloud processing increment range includes:

[0015] Update the point cloud processing range to be tracked and recognized according to the following formula:

[0016] P = P a + P δ

[0017] where P is the point cloud processing range to be tracked and recognized, P a is the current-frame point cloud processing range, and P δ is the next-frame point cloud processing increment range.

[0018] Further, the determining the next-frame point cloud processing increment range according to the next-frame point cloud target tracking and recognition parameters includes:

[0019] Determine the next-frame point cloud processing increment range according to the distance, as well as the maximum and minimum values of the current-frame point cloud target in each coordinate component. Specifically, it includes:

[0020] Determine the next-frame point cloud processing increment range P δ :

[0021]

[0022] Among them, δ x , δ y and δ z are the incremental changes of P δ in the x, y, and z directions respectively; μ x , μ y , μ z ∈(-1, 1) represents the scaling transformation coefficients of P δ in the x, y, and z directions. μ x , μ y , μ z change in positive correlation according to the said distance; x Paxmax and x Paxmin are the maximum and minimum coordinate values of the current frame point cloud target in the x direction, x Paymax and x Paymin are the maximum and minimum coordinate values of the current frame point cloud target in the y direction, x Pazmax and x Pazmin are the maximum and minimum coordinate values of the current frame point cloud target in the z direction.

[0023] Furthermore, the μ x , μ y , μ z are calculated according to the following formula:

[0024]

[0025] Among them, g1, g2, g3 are direct proportional functions related to the said distance; L(p o ) is the said distance.

[0026] Furthermore, determining the next frame point cloud processing increment range according to the next frame point cloud target tracking and recognition parameters includes:

[0027] Determining the next frame point cloud processing increment range according to the said ratio, specifically including;

[0028] Determining the next frame point cloud processing increment range P δ according to the following formula:

[0029]

[0030] Among them, δ x , δ y and δ z are the incremental changes of P δ in the x, y, and z directions respectively; λ represents the size control factor of P δ ; N(P a ) is the current frame point cloud data volume, and N(P) is the background point cloud data volume; and are variation functions corresponding to the x, y, and z directions respectively with as the independent variable.

[0031] Furthermore, the value of λ is calculated according to the following formula:

[0032]

[0033] where a1 is the first preset percentage threshold and a2 is the second preset percentage threshold.

[0034] Furthermore, determining the next-frame point cloud processing increment range according to the next-frame point cloud target tracking and recognition parameters includes:

[0035] Determining the next-frame point cloud processing increment range according to the distance, the maximum and minimum values of the current-frame point cloud target in each coordinate component, and the ratio; specifically including:

[0036] Determining the next-frame point cloud processing increment range P according to the following formula δ :

[0037]

[0038] where δ x , δ y , and δ z are the incremental changes of P δ in the x, y, and z directions respectively; λ represents the size control factor of P δ ; N(P a ) is the current-frame point cloud data volume and N(P) is the background point cloud data volume; and are variation functions corresponding to the x, y, and z directions respectively with as the independent variable; μ x , μ y , μ z ∈(-1,1) represents the scaling transformation coefficients of P δ in the x, y, and z directions, and μ x , μ y , μ z change in positive correlation according to the distance; x Paxmax and x Paxmin are the maximum and minimum coordinate values of the current-frame point cloud target in the x direction, x Paymax and x Paymin are the maximum and minimum coordinate values of the current-frame point cloud target in the y direction, x Pazmax and x Pazmin are the maximum and minimum coordinate values of the current-frame point cloud target in the z direction.

[0039] Further, tracking and recognizing the point cloud target in the processing range of the point cloud to be tracked and recognized to obtain the tracking and recognition result of the point cloud target includes:

[0040] Tracking and recognizing the point cloud target in the processing range of the point cloud to be tracked and recognized based on the target detection algorithm to obtain the tracking and recognition result of the point cloud target.

[0041] The present invention discloses a lidar that executes the above point cloud target tracking and recognition method.

[0042] Compared with the prior art, the present invention can at least achieve one of the following beneficial effects:

[0043] For the point cloud target tracking and recognition method provided by the present invention, the tracking and recognition parameters of the point cloud target in the next frame are determined according to the tracking and recognition result of the point cloud target in the current frame, and the increment range of the point cloud processing in the next frame is determined according to the tracking and recognition parameters of the point cloud target in the next frame; the processing range of the point cloud to be tracked and recognized is updated according to the current frame point cloud processing range and the next frame point cloud processing increment range; the point cloud target in the processing range of the point cloud to be tracked and recognized is tracked and recognized to obtain the point cloud tracking and recognition result. The data processing amount of the point cloud target tracking and recognition is reduced, and the efficiency of the point cloud target tracking and recognition is improved.

[0044] For the lidar provided by the present invention, which executes the point cloud target tracking and recognition method, the tracking and recognition parameters of the point cloud target in the next frame are determined according to the tracking and recognition result of the point cloud target in the current frame, and the increment range of the point cloud processing in the next frame is determined according to the tracking and recognition parameters of the point cloud target in the next frame; the processing range of the point cloud to be tracked and recognized is updated according to the current frame point cloud processing range and the next frame point cloud processing increment range; the point cloud target in the processing range of the point cloud to be tracked and recognized is tracked and recognized to obtain the point cloud tracking and recognition result. The data processing amount of the lidar is reduced, and the efficiency of the lidar point cloud tracking and recognition is improved.

[0045] In the present invention, the above technical solutions can also be combined with each other to achieve more preferred combination schemes. Other features and advantages of the present invention will be described in the following specification, and some advantages can be made obvious from the specification, or understood by implementing the present invention. The purpose and other advantages of the present invention can be achieved and obtained through the content specifically pointed out in the specification and the drawings. Description of the Drawings

[0046] The drawings are only for the purpose of showing specific embodiments and are not considered to be a limitation of the present invention. Throughout the drawings, the same reference signs represent the same components.

[0047] Figure 1 It is a flowchart of the point cloud target tracking and recognition method in the embodiment of the present invention;

[0048] Figure 2 Flow chart of the point cloud target tracking and recognition method in another embodiment of the present invention;

[0049] Figure 3 Schematic diagram of the spatial position relationship between the lidar point cloud and the bounding box in the embodiment of the present invention. Detailed implementation manners

[0050] The following will specifically describe the preferred embodiments of the present invention with reference to the accompanying drawings. The accompanying drawings form a part of this application and are used together with the embodiments of the present invention to explain the principles of the present invention, rather than to limit the scope of the present invention.

[0051] A specific embodiment of the present invention discloses a point cloud target tracking and recognition method. The flow chart is as Figure 1 shown and includes the following steps:

[0052] Step S1: Determine the next-frame point cloud target tracking and recognition parameters according to the current-frame point cloud target tracking and recognition result, and determine the next-frame point cloud processing increment range according to the next-frame point cloud target tracking and recognition parameters.

[0053] Step S2: Update the point cloud processing range to be tracked and recognized according to the current-frame point cloud processing range and the next-frame point cloud processing increment range.

[0054] Step S3: Track and recognize the point cloud target in the point cloud processing range to be tracked and recognized to obtain the point cloud target tracking and recognition result.

[0055] In the above step S1, it should be noted that the next-frame point cloud processing increment range and the next-frame point cloud target tracking and recognition parameters are the same frame, rather than the next frame after the next-frame point cloud target tracking and recognition parameters. As Figure 2 shown, the target detection algorithm for tracking and recognizing the current-frame point cloud target can be determined. The target detection algorithm can include a tracking algorithm based on mean shift, a tracking algorithm based on Kalman filtering, a tracking algorithm based on deep learning, etc. Determining the target detection algorithm according to the point cloud monitoring and tracking data is a mature technology and will not be elaborated here.

[0056] By executing the target detection algorithm in the lidar, the current-frame point cloud target is tracked and recognized. The recognition result includes the detection result of the current-frame point cloud target, and can further include the distance between the above point cloud target and the lidar, the maximum and minimum values of the current-frame point cloud target in each coordinate component, that is, the maximum coordinate value and the minimum coordinate value of the current-frame point cloud target in the x direction, the maximum coordinate value and the minimum coordinate value in the y direction, and the maximum coordinate value and the minimum coordinate value in the z direction, that is, the above six coordinate values.

[0057] The recognition result may further include: the ratio between the current frame point cloud data volume and the background point cloud data volume. The current frame point cloud data volume is the current frame point cloud data volume corresponding to the above-mentioned point cloud target. Usually, the point cloud target only occupies a part in the frame image, and the other part is used as the background, and the corresponding point cloud data volume is the background point cloud data volume.

[0058] Further, determining the next frame point cloud target tracking and recognition parameters according to the current frame point cloud target tracking and recognition result includes:

[0059] Determining the distance between the current frame point cloud target and the lidar, and taking the distance, as well as the maximum and minimum values of the current frame point cloud target in each coordinate component, as the next frame point cloud target tracking and recognition parameters;

[0060] And / or, determining the current frame point cloud data volume corresponding to the current frame point cloud target, and taking the ratio between the current frame point cloud data volume and the background point cloud data volume as the next frame point cloud target tracking and recognition parameter. That is, the above distance and the maximum and minimum values of the current frame point cloud in each coordinate; and / or the above ratio can be used as the next frame point cloud target tracking and recognition parameters.

[0061] Further, determining the next frame point cloud processing increment range according to the next frame point cloud target tracking and recognition parameter includes:

[0062] Determining the next frame point cloud processing increment range according to the distance, as well as the maximum and minimum values of the current frame point cloud target in each coordinate component, specifically including:

[0063] Determining the next frame point cloud processing increment range P according to the following formula δ :

[0064]

[0065] where, δ x , δ y and δ z are the incremental changes of P δ in the x, y, and z directions respectively; μ x , μ y , μ z ∈(-1, 1) represents the scaling transformation coefficients of P δ in the x, y, and z directions, and μ x , μ y , μ z change positively according to the distance; x Paxmax and x Paxmin are the maximum and minimum coordinate values of the current frame point cloud target in the x direction, x Paymax and x PayminThe maximum coordinate value and the minimum coordinate value of the current frame point cloud target in the y direction, x Pazmax and x Pazmin are the maximum coordinate value and the minimum coordinate value of the current frame point cloud target in the z direction. As Figure 3 shown, the processing increment range P of the next frame of point cloud δ can be represented by a bounding box.

[0066] P δ The value of P can be positive or negative. Since the point cloud target is moving, if the current frame point cloud processing range is used as the next frame's point cloud processing range, the point cloud target will not be detected. Therefore, it is necessary to continuously update the current frame point cloud processing range. The main basis for the update is the number of features contained in the point cloud target. When the detected point cloud target is far from the lidar, its point cloud will become sparse, and the number of features contained in the point cloud target will decrease.

[0067] If the size of the bounding box continues to remain unchanged, the detection accuracy will decrease. Therefore, the size of the bounding box is related to the distance between the point cloud target detected in the previous frame and the lidar. In addition, when the size of the bounding box is updated, the dimensions of the length, width, and height of the point cloud target are different, so the change amplitude in each direction of the length, width, and height will also be different.

[0068] The δ in this formula x , δ y and δ z can change independently in their respective directions.

[0069] μ x , μ y , μ z The values of μ will continuously change as the point cloud target moves. The greater the distance, the greater their values; the smaller the distance, the smaller their values.

[0070] Furthermore, the μ x , μ y , μ z are calculated according to the following formula:

[0071]

[0072] where g1, g2, g3 are proportional functions related to the distance; L(p o ) is the distance.

[0073] The proportional function is a well-known function. For example, y = kx, where k can be set independently according to the actual situation and can be selected as 1.2. The larger the value of x, the larger the value of y; the smaller the value of x, the smaller the value of y.

[0074] p oIt can be the center of the object point cloud set P a , and its calculation can be obtained by averaging the Euclidean distances of all the extreme values of the point cloud coordinates. For example, for the object point cloud set P a with a total of m point clouds, their coordinates are respectively denoted as (x1, y1, z1)…(xm, ym, zm), and the center coordinates of the object point cloud set P a are obtained through the following formula:

[0075] ((x1, y1, z1)+…+(xm, ym, zm)) / m.

[0076] That is, the farther the point cloud target to be detected is, the sparser its point cloud will become, and the fewer features regarding the point cloud target to be detected will be. Therefore, the bounding box needs to be appropriately enlarged to prevent missed detection. When the point cloud target to be detected is relatively close, the details and features of the feature points can be well represented. Therefore, the size of the bounding box can be appropriately reduced to reduce the computational workload.

[0077] Furthermore, determining the next-frame point cloud processing increment range according to the next-frame point cloud target tracking and recognition parameters includes:

[0078] Determining the next-frame point cloud processing increment range according to the proportion, specifically including;

[0079] Determining the next-frame point cloud processing increment range P according to the following formula δ :

[0080]

[0081] where δ x , δ y and δ z are the incremental changes of P δ in the x, y, and z directions respectively; λ represents the size control factor of P δ ; N(P a ) is the current-frame point cloud data volume, and N(P) is the background point cloud data volume; and are the change functions corresponding to the x, y, and z directions respectively with as the independent variable. Further, this change function can be a linear function, and the linear function is, for example When , When ,

[0082] Furthermore, the value of λ is calculated according to the following formula:

[0083]

[0084] Among them, a1 is the first preset percentage threshold, and a2 is the second preset percentage threshold. a1 can be selected as 30%; a2 can be selected as 70%.

[0085] By expanding or shrinking the bounding box, the proportion of the point cloud data volume of the current frame is continuously controlled, while reducing the data volume, the accuracy of point cloud object detection is also ensured.

[0086] Further, determining the next-frame point cloud processing increment range according to the next-frame point cloud object tracking and recognition parameters includes:

[0087] Determining the next-frame point cloud processing increment range according to the distance, the maximum and minimum values of the current-frame point cloud object in each coordinate component, and the proportion; specifically including:

[0088] Determine the next-frame point cloud processing increment range P according to the following formula δ :

[0089]

[0090] Among them, δ x , δ y and δ z are the incremental changes of P δ in the x, y, and z directions respectively; λ represents the size control factor of P δ ; N(P a ) is the current-frame point cloud data volume, and N(P) is the background point cloud data volume; and are the change functions corresponding to the x, y, and z directions respectively with as the independent variable; μ x , μ y , μ z ∈(-1, 1) represents the stretching and shrinking transformation coefficients of P δ in the x, y, and z directions, and μ x , μ y , μ z change positively according to the distance; x Paxmax and x Paxmin are the maximum and minimum coordinate values of the current-frame point cloud object in the x direction, x Paymax and x Paymin are the maximum and minimum coordinate values of the current-frame point cloud object in the y direction, x Pazmax and x Pazminare the maximum coordinate value and the minimum coordinate value of the current frame point cloud target in the z direction. That is, by combining the influence of the distance between the current frame point cloud target and the lidar on the processing increment range of the next frame of point cloud, and the influence of the ratio between the current frame point cloud data volume and the background point cloud data volume on the processing increment range of the next frame of point cloud, the value of the processing increment range of the next frame of point cloud can be made more reasonable.

[0091] Among them, the calculation of the value of λ can refer to the above description.

[0092] In the above step 2, further, the updating of the point cloud processing range to be tracked and recognized according to the current frame point cloud processing range and the processing increment range of the next frame of point cloud includes:

[0093] Update the point cloud processing range to be tracked and recognized according to the following formula:

[0094] P = P a + P δ

[0095] where P is the point cloud processing range to be tracked and recognized, P a is the current frame point cloud processing range, and P δ is the processing increment range of the next frame of point cloud. Referring to Figure 2 , each time the processing increment range P δ (point cloud bounding box increment) of the next frame of point cloud is calculated, the value of P changes once. P represents that the range fed into the target detection algorithm or network F each time can be dynamically changed in real time, which is different from the existing method where the point cloud input each time is always all the point clouds in the field of view.

[0096] The construction of the point cloud processing range to be tracked and recognized may include the following steps:

[0097] The initial P is all the point clouds collected in the current frame. Thus, the object point cloud set P a detected in the current frame can be obtained, that is, P a = F(P). For the next frame of point cloud f n+1 input next, within the limited data range detected by the lidar, within a very small time interval, the position of the point cloud target will not undergo a sudden change in position. It can be considered that the distance between the position of the target point cloud set in the next frame of point cloud and the position of the point cloud target in the current frame changes little. That is, the point cloud around the point cloud set P a detected in the Nth frame of point cloud is used as the input point cloud set P for the next frame of detection.

[0098] In the above step S3, further, the tracking and recognition of the point cloud target in the point cloud processing range to be tracked and recognized to obtain the point cloud target tracking and recognition result includes:

[0099] Based on the object detection algorithm, track and identify the point cloud objects in the point cloud processing range to be tracked and identified, and obtain the point cloud object tracking and identification results. That is, every time the point cloud in the point cloud processing range to be tracked and identified is tracked and identified, it can be based on the object detection algorithm.

[0100] After this step is completed, it can continue to return to execute the above step S1, that is, periodically execute determining the tracking and identification parameters of the next frame of point cloud object according to the current frame of point cloud object tracking and identification results, and subsequent steps. And in each execution process, the point cloud processing range to be tracked and identified is updated according to the current frame of point cloud processing range and the point cloud processing increment range of the next frame.

[0101] Further, before executing the above step S1, the following steps may also be included:

[0102] Preprocess the data collected by the lidar to obtain a detection method F that can achieve point cloud object detection.

[0103] For the Nth frame of input point cloud f to be detected n , first perform certain preprocessing on it to facilitate subsequent point cloud object detection.

[0104] The preprocessing mainly includes point cloud smoothing, outlier removal, and downsampling, etc. Point cloud smoothing is to pull the points offset due to errors in the point cloud data back to their original positions, so that the surface of the scanned point cloud object is as close as possible to the surface of the real object. And outlier removal is to directly remove the point cloud with too large offset to reduce its influence on the subsequent point cloud processing. Downsampling is to reduce the amount of point cloud data while maintaining the point cloud features, so as to reduce the data calculation amount while ensuring the detection accuracy.

[0105] Compared with the prior art, the point cloud object tracking and identification method provided by the present invention determines the tracking and identification parameters of the next frame of point cloud object according to the current frame of point cloud object tracking and identification results, and determines the point cloud processing increment range of the next frame according to the tracking and identification parameters of the next frame of point cloud object; updates the point cloud processing range to be tracked and identified according to the current frame of point cloud processing range and the point cloud processing increment range of the next frame; tracks and identifies the point cloud objects in the point cloud processing range to be tracked and identified to obtain the point cloud tracking and identification results. It reduces the data processing amount of point cloud object tracking and identification and improves the efficiency of point cloud object tracking and identification.

[0106] A specific embodiment of the present invention discloses a lidar, which can execute the above point cloud object tracking and identification method. The embodiment of the above point cloud object tracking and identification method can be referred to for description and will not be repeated here.

[0107] The lidar provided by the present invention executes a point cloud target tracking and recognition method, determines the next-frame point cloud target tracking and recognition parameters according to the current-frame point cloud target tracking and recognition result, and determines the next-frame point cloud processing increment range according to the next-frame point cloud target tracking and recognition parameters; updates the point cloud processing range to be tracked and recognized according to the current-frame point cloud processing range and the next-frame point cloud processing increment range; performs tracking and recognition on the point cloud target in the point cloud processing range to be tracked and recognized, and obtains a point cloud tracking and recognition result. This reduces the data processing volume of point cloud target tracking and recognition and improves the efficiency of point cloud target tracking and recognition.

[0108] Those skilled in the art can understand that all or part of the processes for implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a magnetic disk, an optical disk, a read-only memory, or a random access memory, etc.

[0109] As mentioned above, only the preferred specific embodiments of the present invention are described, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for point cloud target tracking and recognition, characterized in that, Including: Determine the next-frame point cloud target tracking and recognition parameters according to the current-frame point cloud target tracking and recognition result, and determine the next-frame point cloud processing increment range according to the next-frame point cloud target tracking and recognition parameters; Update the point cloud processing range to be tracked and recognized according to the current-frame point cloud processing range and the next-frame point cloud processing increment range; Track and recognize the point cloud target in the point cloud processing range to be tracked and recognized, and obtain the point cloud target tracking and recognition result; The determining the next-frame point cloud target tracking and recognition parameters according to the current-frame point cloud target tracking and recognition result includes: Determine the distance between the current-frame point cloud target and the lidar, and use the distance and the maximum and minimum values of the current-frame point cloud target in each coordinate component as the next-frame point cloud target tracking and recognition parameters; And / or, determine the current-frame point cloud data volume corresponding to the current-frame point cloud target, and use the ratio between the current-frame point cloud data volume and the background point cloud data volume as the next-frame point cloud target tracking and recognition parameters; The determining the next-frame point cloud processing increment range according to the next-frame point cloud target tracking and recognition parameters includes: Determine the next-frame point cloud processing increment range according to the distance and the maximum and minimum values of the current-frame point cloud target in each coordinate component, specifically including: Determine the next frame point cloud processing increment range P according to the following formula δ :[[-END]] Among them, δ x , δ y and δ z are the incremental changes of P δ in the x, y, and z directions respectively; μ x , μ y , μ z ∈(-1, 1) represents the scaling transformation coefficients of P δ in the x, y, and z directions, and μ x , μ y , μ z change in positive correlation according to the said distance; x Paxmax and x Paxmin are the maximum and minimum coordinate values of the current frame point cloud target in the x direction, x Paymax and x Paymin are the maximum and minimum coordinate values of the current frame point cloud target in the y direction, x Pazmax and x Pazmin are the maximum and minimum coordinate values of the current frame point cloud target in the z direction; The said μ x , μ y , μ z is calculated according to the following formula: wherein, g1, g2, g3 are direct proportional functions related to the distance; L(p o ) is the distance; The determining the next-frame point cloud processing increment range according to the next-frame point cloud target tracking and recognition parameters includes: Determine the next-frame point cloud processing increment range according to the ratio, specifically including; Determine the next frame point cloud processing increment range P according to the following formula δ : Among them, δ x , δ y and δ z are the incremental changes of P δ in the x, y, and z directions respectively; λ represents the size control factor of P δ ; N(P a ) is the current frame point cloud data volume, and N(P) is the background point cloud data volume; and are change functions corresponding to the x, y, and z directions respectively with as the independent variable; The value of λ is calculated according to the following formula: where a1 is the first preset percentage threshold and a2 is the second preset percentage threshold.

2. The point cloud target tracking and recognition method according to claim 1, wherein The updating the point cloud processing range to be tracked and recognized according to the current-frame point cloud processing range and the next-frame point cloud processing increment range includes: Update the point cloud processing range to be tracked and recognized according to the following formula: P = P a +P δ Among them, P is the processing range of the point cloud to be tracked and recognized, P a is the processing range of the current frame point cloud, P δ is the processing increment range of the next frame point cloud.

3. The point cloud target tracking and recognition method according to claim 1, characterized in that The determining the next-frame point cloud processing increment range according to the next-frame point cloud target tracking and recognition parameters includes: Determine the next-frame point cloud processing increment range according to the distance, the maximum and minimum values of the current-frame point cloud target in each coordinate component, and the ratio; specifically including: Determine the next frame point cloud processing increment range P according to the following formula δ : Among them, δ x , δ y and δ z are the incremental changes of P δ in the x, y, and z directions respectively; λ represents the size control factor of P δ ; N(P a ) is the current frame point cloud data volume, and N(P) is the background point cloud data volume; and are the variation functions corresponding to the x, y, and z directions respectively with as the independent variable; μ x , μ y , μ z ∈(-1, 1) represents the scaling transformation coefficients of P δ in the x, y, and z directions. μ x , μ y , μ z change in positive correlation according to the said distance; x Paxmax and x Paxmin are the maximum and minimum coordinate values of the current frame point cloud target in the x direction, x Paymax and x Paymin are the maximum and minimum coordinate values of the current frame point cloud target in the y direction, x Pazmax and x Pazmin are the maximum and minimum coordinate values of the current frame point cloud target in the z direction.

4. The point cloud target tracking and recognition method according to claim 1, characterized in that The tracking and recognizing the point cloud target in the point cloud processing range to be tracked and recognized to obtain the point cloud target tracking and recognition result includes: Based on the target detection algorithm, track and recognize the point cloud target in the point cloud processing range to be tracked and recognized, and obtain the point cloud target tracking and recognition result.

5. A lidar, characterized in that, Including: Execute the point cloud target tracking and recognition method according to any one of claims 1 to 4.

Images