A method and device for tracking a target object
By using multi-dimensional data space and motion feature prediction technology in radar equipment, the loss problem caused by interference sources during the target object tracking process is solved, and continuous tracking and stability improvement of target objects is achieved.
Patent Information
- Application Number
- CN202411252667.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-09
- Publication Date
- 2025-02-11
- Estimated Expiration
- 2044-09-09
AI Technical Summary
Existing radar equipment is easily affected by interference sources (subsequent rain, sand, thick fog, biological activities, etc.) during the tracking of target objects, resulting in the loss of target objects and the tracking results becoming cluttered.
By obtaining point cloud data continuously detected by radar equipment, it is imported into the multidimensional data space in chronological order, extracting the geometric outline of the target to be identified and matching it with the historical data. If matched, it is marked as the target object; if not matched, the geometric outline of the target object is predicted using motion features and corrected the spatial dataset to achieve continuous tracking of the target object.
It effectively avoids the loss problem caused by interference sources during the target object tracking process, realizes continuous tracking of target objects, and improves the stability and accuracy of tracking.
Smart Images
Figure CN118759489B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of sensor technology, and in particular to a method and device for tracking a target object. Background Art
[0002] With the gradual popularization of lidar and other related radar equipment in the civilian field, the scenarios they have to deal with are gradually increasing, and the challenges they face are becoming more severe, which has greatly exposed some inherent defects of radar equipment.
[0003] In the prior art, radar-type detection equipment generally adopts the detection method of transmitting signal waves; the signal waves spread in a fixed direction; contact the target object; reflect the signal waves; and the radar device captures the reflected signal to calculate the position of the target object.
[0004] In the real world, abnormal interference sources are everywhere, such as rain, dust, fog, biological activities, etc. Once there is an interference source between the radar device and the target object, the radar signal emitted by the radar device is difficult to project onto the correct target object, causing the target object observed by the radar device to be lost. As a result, the output of the radar device becomes clutter, resulting in the sudden loss of the tracking target when tracking the target object. Summary of the invention
[0005] In view of this, an embodiment of the present invention provides a method and device for tracking a target object, so as to achieve the purpose of continuous tracking of the target object.
[0006] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:
[0007] A first aspect of an embodiment of the present invention discloses a method for tracking a target object, the method comprising:
[0008] Acquire point cloud data obtained by radar equipment continuously detecting the target space area;
[0009] Importing the point cloud data at each moment into a preset multidimensional data space in chronological order, to obtain the multidimensional data space containing various spatial data sets; each of the spatial data sets corresponds to the point cloud data at one of the moments;
[0010] For each of the spatial data sets, extract the geometric outline of the target to be identified from the spatial data set, and determine whether the geometric outline matches the historical data; if so, mark the target to be identified as a target object; the historical data includes the geometric outline of the target object identified in the past;
[0011] splicing the spatial data sets in time order to obtain the motion characteristics of the target object;
[0012] For each of the spatial data sets, if the target object is not identified, the geometric contour of the target object in the spatial data set is predicted using the motion feature, and the spatial data set is corrected based on the predicted geometric contour;
[0013] The corrected spatial data set is used to complete the corresponding point cloud data.
[0014] Preferably, the step of importing the point cloud data at each moment into a preset multidimensional data space in chronological order to obtain the multidimensional data space containing each spatial data set includes:
[0015] The point cloud coordinates in the point cloud data at each moment are converted in chronological order to obtain the target space coordinates corresponding to each point cloud coordinate in the multidimensional data space, and the data identifier corresponding to each target space coordinate in the multidimensional data space is modified to 1 to obtain the multidimensional data space containing each spatial data set; the data identifier corresponding to each spatial coordinate in the multidimensional data space is 0 in the initial state.
[0016] Preferably, the method further comprises:
[0017] When, for each of the spatial data sets, it is identified from the spatial data sets that the geometric outline of the target to be identified is incomplete, extracting local features from the geometric outline;
[0018] Based on the local feature, a query is performed in a preset feature database. If the local feature matches a feature in the feature database, the local feature is determined to be a local feature of the target object, and the local feature is marked as a target local feature;
[0019] For each of the spatial data sets, if the target object is not identified but the target local features exist, the motion features and the target local features are used to predict the geometric contour of the target object in the spatial data set, and the spatial data set is corrected based on the predicted geometric contour.
[0020] Preferably, the method further comprises:
[0021] If the geometrical outline of the target to be identified does not match the historical data, marking the target to be identified as an auxiliary object;
[0022] When the target object is not identified in any of the spatial data sets corresponding to multiple consecutive moments, the spatial data sets are spliced in time sequence to obtain the motion features of the auxiliary object;
[0023] Based on the motion characteristics of the auxiliary object, identifying the object type of the auxiliary object;
[0024] Querying the linkage relationship between the auxiliary object and the target object based on the object type;
[0025] The linkage relationship is used to predict the geometric contours of the target object in the spatial data set corresponding to each continuous moment, and the predicted geometric contours are used to correct the spatial data set corresponding to each continuous moment.
[0026] Preferably, the method further comprises:
[0027] If the object type of the auxiliary object is a natural interference type, after the corresponding point cloud data is completed using the corrected spatial data set, the point cloud data is filtered to remove the point cloud data corresponding to the auxiliary object.
[0028] A second aspect of an embodiment of the present invention discloses a device for tracking a target object, the device comprising:
[0029] An acquisition unit, used to acquire point cloud data obtained by the radar device continuously detecting the target space area;
[0030] An importing unit, used to import the point cloud data at each moment into a preset multidimensional data space in chronological order, to obtain the multidimensional data space containing various spatial data sets; each of the spatial data sets corresponds to the point cloud data at one of the moments;
[0031] an identification unit, for extracting, for each of the spatial data sets, a geometric outline of a target to be identified from the spatial data set, and determining whether the geometric outline matches historical data; if so, marking the target to be identified as a target object; the historical data includes geometric outlines of the target objects identified in the past;
[0032] A splicing unit, used for splicing the spatial data sets in time order to obtain the motion characteristics of the target object;
[0033] A first prediction unit is configured to, for each of the spatial data sets, predict a geometric contour of the target object in the spatial data set using the motion feature if the target object is not identified, and correct the spatial data set based on the predicted geometric contour;
[0034] A completion unit is used to complete the corresponding point cloud data using the corrected spatial data set.
[0035] Preferably, the introduction unit is specifically used for:
[0036] The point cloud coordinates in the point cloud data at each moment are converted in chronological order to obtain the target space coordinates corresponding to each point cloud coordinate in the multidimensional data space, and the data identifier corresponding to each target space coordinate in the multidimensional data space is modified to 1 to obtain the multidimensional data space containing each spatial data set; the data identifier corresponding to each spatial coordinate in the multidimensional data space is 0 in the initial state.
[0037] Preferably, the device further comprises:
[0038] The second prediction unit is used to extract local features from the geometric outline of the target to be identified in the spatial data set when the geometric outline of the target to be identified is incomplete for each of the spatial data sets; query a preset feature database based on the local features, and if the local features match the features in the feature database, determine that the local features are local features of the target object, and mark the local features as target local features; for each of the spatial data sets, if the target object is not identified but the target local features exist, use the motion features and the target local features to predict the geometric outline of the target object in the spatial data set, and correct the spatial data set based on the predicted geometric outline.
[0039] Preferably, the device further comprises:
[0040] The third prediction unit is used to mark the target to be identified as an auxiliary object if the geometric contour of the target to be identified does not match the historical data; when the target object is not identified in the spatial data sets corresponding to multiple consecutive moments, splice the various spatial data sets in chronological order to obtain the motion characteristics of the auxiliary object; identify the object type of the auxiliary object based on the motion characteristics of the auxiliary object; query the linkage relationship between the auxiliary object and the target object based on the object type; use the linkage relationship to predict the geometric contour of the target object in the spatial data sets corresponding to each consecutive moment, and use the predicted geometric contour to correct the spatial data sets corresponding to each consecutive moment.
[0041] Preferably, the device further comprises:
[0042] A filtering unit is used for filtering the point cloud data after completing the corresponding point cloud data with the corrected spatial data set if the object type of the auxiliary object is a natural interference type, so as to eliminate the point cloud data corresponding to the auxiliary object.
[0043] Based on a target object tracking method and device provided by the above-mentioned embodiment of the present invention, point cloud data obtained by a radar device continuously detecting a target space area is obtained; the point cloud data at each moment is sequentially imported into a preset multidimensional data space in chronological order to obtain the multidimensional data space containing various spatial data sets; each of the spatial data sets corresponds to the point cloud data at one of the moments; for each of the spatial data sets, a geometric contour of a target to be identified is extracted from the spatial data set, and it is determined whether the geometric contour matches historical data; if so, the target to be identified is marked as a target object; the historical data includes the geometric contours of the target objects identified in the past; the various spatial data sets are spliced in chronological order to obtain motion features of the target object; for each of the spatial data sets, if the target object is not identified, the motion features are used to predict the geometric contour of the target object in the spatial data set, and the spatial data set is corrected based on the predicted geometric contour; the corrected spatial data set is used to complete the corresponding point cloud data. In this scheme, the point cloud data sent by the radar is reconstructed multiple times in the time dimension to obtain the motion characteristics of the target object. When the radar detection of the target object is disturbed and the point cloud data of the target object is missing at a certain moment, the motion characteristics are used to predict the geometric contour of the target object and complete the point cloud data to achieve continuity in tracking the target object. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0045] Figure 1 A flow chart of a target object tracking method disclosed in an embodiment of the present invention;
[0046] Figure 2 The present invention is a structural diagram of a target object tracking device disclosed in an embodiment of the present invention. DETAILED DESCRIPTION
[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0048] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.
[0049] As can be seen from the background technology, if there is an interference source between the radar device and the target object, the radar signal emitted by the radar device is difficult to project onto the correct target object, causing the target object observed by the radar device to be lost. As a result, the output of the radar device becomes clutter, resulting in the sudden loss of the tracking target when tracking the target object.
[0050] Therefore, the embodiment of the present invention discloses a method and device for tracking a target object. In this solution, the point cloud data sent by the radar is reconstructed multiple times in the time dimension to obtain the motion characteristics of the target object. When the radar detection of the target object is disturbed and the point cloud data of the target object is missing at a certain moment, the motion characteristics are used to predict the geometric contour of the target object and complete the point cloud data to achieve the continuity of tracking the target object.
[0051] like Figure 1 FIG. 1 is a flow chart of a method for tracking a target object disclosed in an embodiment of the present invention, which mainly includes the following steps:
[0052] Step S101: Acquire point cloud data obtained by the radar device continuously detecting the target space area.
[0053] In step S101, the target space area refers to a production environment, such as a site where coal is loaded into a train compartment.
[0054] Point cloud data is obtained by capturing the reflected signals of various objects in the target space area after the radar device transmits the radar signal. The point cloud data includes the point cloud coordinates corresponding to each data point to indicate the position of the data point in space.
[0055] It is understandable that it is necessary to ensure that complete point cloud data is detected at least in a period of time, so that this period of point cloud data can be used to subsequently identify the motion characteristics of the target object in the target space area.
[0056] It should be noted that radar equipment can be installed at multiple different angles to observe the same target in the same target space area. This can avoid some one-way interference sources and can also be used for double correction of radar signals (i.e. point cloud data) when large-scale interference occurs, thereby improving accuracy.
[0057] Step S102: importing the point cloud data at each moment into a preset multidimensional data space in chronological order, to obtain a multidimensional data space containing various spatial data sets.
[0058] Each spatial data set corresponds to point cloud data at a certain moment. Multidimensional data space adds a time dimension to the three-dimensional data space.
[0059] In the specific implementation process of step S102, each point cloud coordinate in the point cloud data at each moment is converted in chronological order to obtain the target space coordinates corresponding to each point cloud coordinate in the multidimensional data space, and the data identifier corresponding to each target space coordinate in the multidimensional data space is modified to 1 to obtain a multidimensional data space containing each spatial data set.
[0060] The data identifier corresponding to each spatial coordinate in the multidimensional data space is 0 in the initial state.
[0061] It should be noted that radar signals are imported into the multidimensional data space in the form of point cloud or other similar formats, and each data point of the point cloud data is identified in the multidimensional data space coordinate system in the form of three-dimensional spatial coordinates. The coordinates corresponding to the data points are identified as 1, and other positions are identified as 0, thus forming a set of binary data sets (i.e., spatial data sets) alternating between 0 and 1. A spatial data set identifies the instantaneous status of all observable targets in the target space area at a time node.
[0062] Step S103: for each spatial data set, extract the geometric outline of the target to be identified from the spatial data set, and determine whether the geometric outline matches the historical data; if so, mark the target to be identified as a target object.
[0063] In step S103, the contours enclosed by the spatial coordinates with data identification of 1 pair in the spatial data set are the geometric contours of the target to be identified. The historical data includes the geometric contours of the target objects identified in the past.
[0064] It should be noted that in most production environments, the main target objects are usually production targets including: mechanical targets and material targets controlled by them.
[0065] For example, the historical data includes a previously identified train carriage outline (ie, target object). When the geometric outline of the target to be identified matches the train carriage outline of the historical data, the target to be identified is determined to be the train carriage outline.
[0066] Preferably, a match is determined when the similarity between the geometric outline of the target to be identified and the train carriage outline of the historical data reaches a set threshold.
[0067] Step S104: splicing the various spatial data sets in time order to obtain the motion features of the target object.
[0068] In step S104, the spatial data sets at different time nodes are spliced in the time dimension, and the motion characteristics such as displacement direction, speed, and morphological change of the target object can be estimated according to the change of the position of the target object in each spatial data set.
[0069] It can be understood that when detecting point cloud data, it is ensured that the complete point cloud data of the target object is identified within at least a time segment. Therefore, after combining the spatial data sets corresponding to these point cloud data, the motion characteristics of the target object within this time segment can be identified, and then the motion characteristics can be used to predict the geometric outline of the target object in the spatial data sets before and after this time segment.
[0070] Among them, the closer the time point corresponding to the spatial data set is to this time segment, the higher the accuracy of the prediction will be.
[0071] In the specific implementation process of step S104, only the spatial data sets that continuously and uninterruptedly identify the target object within a certain period of time may be spliced to obtain the motion characteristics of the target object.
[0072] Step S105: For each spatial data set, if the target object is not identified, the geometric contour of the target object in the spatial data set is predicted using the motion features, and the spatial data set is corrected based on the predicted geometric contour.
[0073] In step S105, the spatial data set is corrected based on the predicted geometric contour. Specifically, the data identifier of the spatial coordinate corresponding to the predicted geometric contour in the spatial data set is changed from 0 to 1.
[0074] Exemplarily, the motion features include the displacement direction and displacement speed of the target object. The motion features are combined with the time points corresponding to the spatial data set in which the target object is not identified to estimate the position where the target object appears in the spatial data set, thereby predicting the geometric outline of the target object in the spatial data set.
[0075] In one embodiment, there are some spatial data sets where the geometric contours of the target to be identified are incomplete, and it is impossible to distinguish whether it is the target object based on historical data. In this case, local features can be extracted from the incomplete geometric contours, and the feature database can be used to determine whether it is the local feature of the target object. If the motion features and local features can be combined to predict the complete geometric contour of the target object, the accuracy is higher than that of directly using the motion features for prediction. The specific process is as follows:
[0076] Step S201: when it is determined that the geometric outline of the target to be identified is incomplete for each spatial data set, local features are extracted from the geometric outline.
[0077] For example, in the case of car identification, its complete geometric outline is a cuboid, but because of incomplete point cloud data, only the geometric outline of one of the car corners may be captured, and local features can be extracted from the geometric outline of the car corner. Local features, for example, height features based on the position coordinates of the geometric outline of the car corner; shape features of the car corner, etc.
[0078] Step S202: querying a preset feature database based on the local feature, if the local feature matches the feature in the feature database, determining that the local feature is a local feature of the target object, and marking the local feature as a target local feature.
[0079] In step S202, a plurality of local features of the target object are pre-stored in the feature database.
[0080] For example, if the target object is a carriage, the feature database stores the height features of the carriage, the shape features of each part of the carriage, etc.
[0081] Step S203: For each spatial data set, if the target object is not identified but the target local features exist, the motion features and the target local features are used to predict the geometric contour of the target object in the spatial data set, and the spatial data set is corrected based on the predicted geometric contour.
[0082] In step S203, the geometric outline of the target to be identified in the spatial data set is highly incomplete, resulting in the inability to identify the target object based on the historical data.
[0083] The spatial data set is corrected based on the predicted geometric contour, specifically, the data identifier of the spatial coordinate corresponding to the predicted geometric contour in the spatial data set is changed from 0 to 1.
[0084] In one embodiment, there are multiple consecutive moments in which the target object is not identified in the spatial data set corresponding to each other, indicating that the radar signal is continuously interfered with and the point cloud data is missing for a period of time. In this case, the accuracy of predicting the geometric contour in the spatial data set corresponding to this period of time using only the motion characteristics of the target object is too poor. At this time, it is necessary to combine the auxiliary object for prediction, as follows:
[0085] Step S301: If the geometric outline of the target to be identified does not match the historical data, the target to be identified is marked as an auxiliary object.
[0086] In step S301, the geometric outline of the target to be identified does not match the historical data, which is generally a non-production target, including: biological targets, background targets, natural interference targets, etc.
[0087] For example, water vapor or smoke emitted by coal will quickly drift to the upper area of the carriage, and this drift is often irregular, that is, the displacement direction, speed, and shape changes are relatively random. Such targets are called natural interference targets.
[0088] Although the geometric contours of the water vapor or smoke emitted by the coal are different in each spatial data set, they all tend to gather above the carriage. Therefore, they can assist in predicting the geometric contours of the carriage and mark it as an auxiliary target.
[0089] Step S302: When the target object is not identified in any of the spatial data sets corresponding to multiple consecutive moments, the spatial data sets are spliced in time sequence to obtain motion features of the auxiliary object.
[0090] In step S302, the target object is not recognized in the spatial data sets corresponding to multiple consecutive moments. This may be because the target object is partially or completely blocked during this period, so the geometric outline in the spatial data set is incomplete or lost, resulting in the inability to recognize the target object.
[0091] The spatial data sets at different time nodes are spliced in the time dimension, and the motion characteristics such as displacement direction, speed, and morphological change can be estimated based on the changes in the positions of the auxiliary objects in each spatial data set.
[0092] Step S303: Based on the motion characteristics of the auxiliary object, the object type of the auxiliary object is identified.
[0093] For example, the movement characteristics of the water vapor or smoke emitted by the coal, such as the displacement direction, speed, and shape change, are relatively random, and the object type can be identified as a natural interference type.
[0094] For another example, although the moving direction and speed of some auxiliary objects are not fixed, their shapes are relatively fixed. Such auxiliary objects are usually biological objects.
[0095] For another example, some auxiliary objects have motion characteristics that keep their spatial positions unchanged in the time dimension. This type of auxiliary object is called a background target.
[0096] Step S304: query the linkage relationship between the auxiliary object and the target object based on the object type.
[0097] In step S304, the linkage relationship between the auxiliary object and the target object may be preset.
[0098] For example, the linkage relationship can reflect that the water vapor or smoke (i.e., the auxiliary object) emitted by the coal is directly below the carriage (the target object), that is, the starting position of the water vapor or smoke can be considered as the upper contour of the carriage.
[0099] Step S305: using the linkage relationship to predict the geometric contours of the target object in the spatial data set corresponding to each continuous moment, and using the predicted geometric contours to correct the spatial data set corresponding to each continuous moment.
[0100] In step S305, for a spatial data set having local features of a target object, the linkage relationship and the local features may be combined to predict a geometric outline of the target object in the spatial data set.
[0101] In one embodiment, if the object type of the auxiliary object is a natural interference type, after the corresponding point cloud data is completed using the corrected spatial data set, the point cloud data is filtered to remove the point cloud data corresponding to the auxiliary object.
[0102] Step S106: Use the corrected spatial data set to complete the corresponding point cloud data.
[0103] In step S106, the spatial coordinates of the data marked as 1 in the corrected spatial data set are converted into point cloud coordinates of the data points in the point cloud data, so as to obtain complete point cloud data.
[0104] Based on the target object tracking method disclosed in the above-mentioned embodiment of the present invention, in this scheme, the point cloud data sent by the radar is multiply reconstructed in the time dimension to obtain the motion characteristics of the target object. When the radar detection of the target object is disturbed and the point cloud data of the target object is missing at a certain moment, the motion characteristics are used to predict the geometric contour of the target object and complete the point cloud data to achieve continuity in tracking the target object.
[0105] Corresponding to the target object tracking method disclosed in the above embodiment of the present invention, Figure 2 , which is a structural diagram of a target object tracking device disclosed in an embodiment of the present invention, the device includes: an acquisition unit 201, an import unit 202, a recognition unit 203, a splicing unit 204, a first prediction unit 205 and a completion unit 206.
[0106] The acquisition unit 201 is used to acquire point cloud data obtained by the radar device continuously detecting the target space area.
[0107] The import unit 202 is used to import the point cloud data at each moment into a preset multidimensional data space in chronological order to obtain a multidimensional data space containing various spatial data sets; each spatial data set corresponds to the point cloud data at one moment.
[0108] Preferably, the import unit 202 is specifically used for:
[0109] The point cloud coordinates in the point cloud data at each moment are converted in chronological order to obtain the target space coordinates corresponding to each point cloud coordinate in the multidimensional data space, and the data identifier corresponding to each target space coordinate in the multidimensional data space is modified to 1 to obtain a multidimensional data space containing each spatial data set; the data identifier corresponding to each space coordinate in the multidimensional data space is 0 in the initial state.
[0110] The identification unit 203 is used to extract the geometric outline of the target to be identified from the spatial data set for each spatial data set, and determine whether the geometric outline matches the historical data; if so, mark the target to be identified as a target object; the historical data includes the geometric outlines of the target objects identified in the past.
[0111] The splicing unit 204 is used to splice the various spatial data sets in time sequence to obtain the motion characteristics of the target object.
[0112] The first prediction unit 205 is configured to predict, for each spatial data set, a geometric outline of the target object in the spatial data set using motion features if the target object is not identified, and correct the spatial data set based on the predicted geometric outline.
[0113] The completion unit 206 is used to complete the corresponding point cloud data using the corrected spatial data set.
[0114] Preferably, the tracking device for the target object further comprises:
[0115] The second prediction unit is used to extract local features from the geometric outline of the target to be identified in the spatial data set when the geometric outline of the target to be identified is incomplete for each spatial data set; query a preset feature database based on the local features, and if the local features match the features in the feature database, determine that the local features are local features of the target object, and mark the local features as target local features; for each spatial data set, if the target object is not identified but the target local features exist, use the motion features and the target local features to predict the geometric outline of the target object in the spatial data set, and correct the spatial data set based on the predicted geometric outline.
[0116] Preferably, the tracking device for the target object further comprises:
[0117] The third prediction unit is used to mark the target to be identified as an auxiliary object if the geometric contour of the target does not match the historical data; when the target object is not identified in the spatial data sets corresponding to multiple consecutive moments, the spatial data sets are spliced in chronological order to obtain the motion characteristics of the auxiliary object; based on the motion characteristics of the auxiliary object, the object type of the auxiliary object is identified; based on the object type, the linkage relationship between the auxiliary object and the target object is queried; the geometric contour of the target object in the spatial data sets corresponding to each consecutive moment is predicted by using the linkage relationship, and the spatial data sets corresponding to each consecutive moment are corrected by using the predicted geometric contour.
[0118] Preferably, the tracking device for the target object further comprises:
[0119] The filtering unit is used to filter the point cloud data after completing the corresponding point cloud data with the corrected spatial data set if the object type of the auxiliary object is a natural interference type, so as to eliminate the point cloud data corresponding to the auxiliary object.
[0120] Based on the target object tracking device disclosed in the above-mentioned embodiment of the present invention, in this scheme, the point cloud data sent by the radar is multiply reconstructed in the time dimension to obtain the motion characteristics of the target object. When the radar detection of the target object is disturbed and the point cloud data of the target object is missing at a certain moment, the motion characteristics are used to predict the geometric contour of the target object and complete the point cloud data to achieve continuity in tracking the target object.
[0121] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.
[0122] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0123] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for tracking a target object, characterized in that: The method comprises: Acquire point cloud data obtained by radar equipment continuously detecting the target space area; Importing the point cloud data at each moment into a preset multidimensional data space in chronological order, to obtain the multidimensional data space containing various spatial data sets; each of the spatial data sets corresponds to the point cloud data at one of the moments; For each of the spatial data sets, extracting a geometric outline of the target to be identified from the spatial data set, and determining whether the geometric outline matches historical data; the historical data includes the geometric outline of the target object identified in the past; If yes, the target to be identified is marked as a target object; each of the spatial data sets is spliced in time order to obtain the motion features of the target object; for each of the spatial data sets, if the target object is not identified, the geometric contour of the target object in the spatial data set is predicted using the motion features, and the spatial data set is corrected based on the predicted geometric contour; If the geometric contour of the target to be identified does not match the historical data, the target to be identified is marked as an auxiliary object; when the target object is not identified in the spatial data sets corresponding to multiple consecutive moments, the spatial data sets are spliced in chronological order to obtain the motion characteristics of the auxiliary object; based on the motion characteristics of the auxiliary object, the object type of the auxiliary object is identified; based on the object type, the linkage relationship between the auxiliary object and the target object is queried; the geometric contour of the target object in the spatial data sets corresponding to each consecutive moment is predicted using the linkage relationship, and the spatial data sets corresponding to each consecutive moment are corrected using the predicted geometric contour; The corrected spatial data set is used to complete the corresponding point cloud data.
2. The method according to claim 1, characterized in that The step of importing the point cloud data at each moment into a preset multidimensional data space in chronological order to obtain the multidimensional data space containing each spatial data set includes: The point cloud coordinates in the point cloud data at each moment are converted in chronological order to obtain the target space coordinates corresponding to each point cloud coordinate in the multidimensional data space, and the data identifier corresponding to each target space coordinate in the multidimensional data space is modified to 1 to obtain the multidimensional data space containing each spatial data set; the data identifier corresponding to each spatial coordinate in the multidimensional data space is 0 in the initial state.
3. The method according to claim 1, characterized in that The method further comprises: When, for each of the spatial data sets, it is identified from the spatial data sets that the geometric outline of the target to be identified is incomplete, extracting local features from the geometric outline; Based on the local feature, a query is performed in a preset feature database. If the local feature matches a feature in the feature database, the local feature is determined to be a local feature of the target object, and the local feature is marked as a target local feature; For each of the spatial data sets, if the target object is not identified but the target local features exist, the motion features and the target local features are used to predict the geometric contour of the target object in the spatial data set, and the spatial data set is corrected based on the predicted geometric contour.
4. The method according to claim 1, characterized in that: The method further comprises: If the object type of the auxiliary object is a natural interference type, after the corresponding point cloud data is completed using the corrected spatial data set, the point cloud data is filtered to remove the point cloud data corresponding to the auxiliary object.
5. A tracking device for a target object, characterized in that: The device comprises: An acquisition unit, used to acquire point cloud data obtained by the radar device continuously detecting the target space area; An importing unit, used to import the point cloud data at each moment into a preset multidimensional data space in chronological order, to obtain the multidimensional data space containing various spatial data sets; each of the spatial data sets corresponds to the point cloud data at one of the moments; an identification unit, for extracting, for each of the spatial data sets, a geometric outline of a target to be identified from the spatial data set, and determining whether the geometric outline matches historical data; if so, marking the target to be identified as a target object; the historical data includes geometric outlines of the target objects identified in the past; A splicing unit, used for splicing the spatial data sets in time order to obtain the motion characteristics of the target object; A first prediction unit is configured to, for each of the spatial data sets, predict a geometric contour of the target object in the spatial data set using the motion feature if the target object is not identified, and correct the spatial data set based on the predicted geometric contour; A third prediction unit is configured to mark the target to be identified as an auxiliary object if the geometric contour of the target to be identified does not match the historical data; when the target object is not identified in the spatial data sets corresponding to multiple consecutive moments, splice the spatial data sets in chronological order to obtain the motion characteristics of the auxiliary object; identify the object type of the auxiliary object based on the motion characteristics of the auxiliary object; query the linkage relationship between the auxiliary object and the target object based on the object type; use the linkage relationship to predict the geometric contour of the target object in the spatial data sets corresponding to each consecutive moment, and use the predicted geometric contour to correct the spatial data sets corresponding to each consecutive moment; A completion unit is used to complete the corresponding point cloud data using the corrected spatial data set.
6. The device according to claim 5, characterized in that The import unit is specifically used for: Converting each point cloud coordinate in the point cloud data at each moment in time order, obtaining the target space coordinate corresponding to each point cloud coordinate in the multidimensional data space, and modifying the data identifier corresponding to each target space coordinate in the multidimensional data space to 1, to obtain the multidimensional data space containing each spatial data set; The data identifier corresponding to each space coordinate in the multi-dimensional data space is 0 in the initial state.
7. The device according to claim 5, characterized in that The device also includes: The second prediction unit is used to extract local features from the geometric outline of the target to be identified in the spatial data set when the geometric outline of the target to be identified is incomplete for each of the spatial data sets; query a preset feature database based on the local features, and if the local features match the features in the feature database, determine that the local features are local features of the target object, and mark the local features as target local features; for each of the spatial data sets, if the target object is not identified but the target local features exist, use the motion features and the target local features to predict the geometric outline of the target object in the spatial data set, and correct the spatial data set based on the predicted geometric outline.
8. The device according to claim 5, characterized in that The device also includes: A filtering unit is used for filtering the point cloud data after completing the corresponding point cloud data with the corrected spatial data set if the object type of the auxiliary object is a natural interference type, so as to eliminate the point cloud data corresponding to the auxiliary object.
Citation Information
Patent Citations
Vehicle re-identification method and device based on laser radar camera
CN114693863A