A point cloud data recognition method and system based on multi-dimensional feature target backtracking

Abstract

This invention relates to the field of point cloud data recognition technology, and in particular to a point cloud data recognition method and system based on multi-dimensional feature target backtracking. The method includes: acquiring radar point cloud data of a target object; obtaining the inherent features of the target object by transmitting the radar point cloud data in different directions; obtaining continuous recognition features based on the performance of the target object's inherent features under different interferences; and identifying the target object using the continuous recognition features. This invention combines the relevant characteristics of lidar equipment and point cloud data processing, and makes various adjustments and optimizations to the feature definition, capture, and recognition of the target object from multiple dimensions, thereby making the target's feature data signal more robust and flexible.

Description

Technical Field

[0001] This invention relates to the field of point cloud data recognition technology, and in particular to a point cloud data recognition method and system based on multi-dimensional feature target backtracking. Background Technology

[0002] As lidar devices become increasingly common in daily life, various problems and limitations arising from their inherent characteristics have begun to plague practitioners in related industries. The most typical problem is their susceptibility to interference from factors such as dust.

[0003] In the use of lidar, the acquisition of targets is mainly reflected in the form of point cloud data. That is, the reflection signal of the target object to the radar wave is converted into a point in the point cloud dataset, and this point contains a set of three-dimensional coordinates in that space. When external factors such as smoke and dust interfere, these abnormal targets will block the actual target object from the radar, feeding back false reflection signals to the radar equipment, thus generating many abnormal points in the point cloud dataset.

[0004] When point cloud data analysis programs analyze these points, it becomes difficult to distinguish the authenticity of the point cloud data. Especially when target identification relies on inherent geometric features, the loss of features due to abnormal interference can easily lead to identification failure, ultimately preventing the program from functioning correctly. Therefore, there is an urgent need for a point cloud data identification method and system based on multi-dimensional feature-based target backtracking. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a point cloud data recognition method and system based on multi-dimensional feature target backtracking.

[0006] Firstly, the present invention provides a point cloud data recognition method based on multi-dimensional feature target backtracking, which adopts the following technical solution:

[0007] A point cloud data recognition method based on multi-dimensional feature target backtracking includes:

[0008] Acquire radar point cloud data of the target object;

[0009] By transmitting radar point cloud data in different directions, the inherent characteristics of the target object can be obtained;

[0010] Based on the inherent characteristics of the target object and their performance under different interferences, continuous identification features are obtained;

[0011] Target objects are identified using continuous recognition features;

[0012] If recognition fails

[0013] The point cloud coordinates of continuously recognizable features in radar point cloud data are obtained by using multi-dimensional spatial backtracking;

[0014] Complete point cloud data is obtained by fine-tuning the point cloud coordinates of continuously identified features.

[0015] Furthermore, the method of obtaining the inherent characteristics of the target by transmitting radar point cloud data in different directions includes setting the direction of the target's movement as the X-axis direction, and then selecting the projection direction as the Y-axis or Z-axis direction.

[0016] Furthermore, the continuous identification features are obtained based on the inherent characteristics of the target object and its performance under different interferences. This includes manually observing and determining whether the projected point cloud has relatively clear identification features, whether the features are stable and continuous, whether they are easily lost due to interference from position or other factors, whether the time of loss is greater than 1 second, and whether the number of times they are lost is greater than 3.

[0017] Furthermore, the method of identifying the target object using continuous identification features includes identifying the target object using an AI model based on the continuous identification features.

[0018] Furthermore, the point cloud coordinates of the continuous identification features of radar point cloud data obtained by using multi-dimensional space backtracking include backtracking the continuous identification features of the target point cloud in the previous time period, mapping the local data of the continuous identification features to the same multi-dimensional data space, and obtaining a time-based continuous target feature point cloud set.

[0019] Furthermore, the method of obtaining the point cloud coordinates of the continuous identification features of radar point cloud data by using multi-dimensional spatial backtracking also includes calculating the velocity, acceleration, and direction of the target feature point cloud in the continuous target feature point cloud set, and then obtaining the position of the target feature point cloud.

[0020] Furthermore, the method of obtaining the point cloud coordinates of the continuous identification features of the radar point cloud data by using multi-dimensional spatial backtracking also includes calibrating the target feature point cloud position and the radar point cloud data, and fine-tuning the target feature point cloud position to obtain the repaired target point cloud and complete the identification.

[0021] Secondly, a point cloud data recognition system based on multi-dimensional feature target backtracking includes:

[0022] The data acquisition module is configured to acquire radar point cloud data of the target object;

[0023] The identification module is configured to obtain the inherent features of the target object by transmitting radar point cloud data in different directions; obtain continuous identification features based on the performance of the inherent features of the target object under different interferences; and identify the target object using the continuous identification features.

[0024] The calculation module is configured to, if recognition fails, use multi-dimensional space backtracking to obtain the point cloud coordinates of the continuously identified features of the radar point cloud data; and obtain complete point cloud data by fine-tuning the point cloud coordinates of the continuously identified features.

[0025] Thirdly, the present invention provides a computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the point cloud data recognition method based on multi-dimensional feature target backtracking.

[0026] Fourthly, the present invention provides a terminal device, including a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor to provide a point cloud data recognition method based on multi-dimensional feature target backtracking.

[0027] In summary, the present invention has the following beneficial technical effects:

[0028] This invention combines the characteristics of lidar equipment and point cloud data processing, making various adjustments and optimizations to the feature definition, capture, and identification of targets from multiple dimensions. This results in more robust and flexible target feature data signals. Furthermore, it endows point cloud data, which originally only had three-dimensional spatial characteristics, with rich features in multiple dimensions, ultimately enabling it to filter and avoid interference signals at a higher dimension. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of a point cloud data recognition method based on multi-dimensional feature target backtracking according to Embodiment 1 of the present invention. Detailed Implementation

[0030] The present invention will be further described in detail below with reference to the accompanying drawings.

[0031] Example 1

[0032] Reference Figure 1 This embodiment of a point cloud data recognition method based on multi-dimensional feature target backtracking includes:

[0033] Acquire radar point cloud data of the target object;

[0034] By transmitting radar point cloud data in different directions, the inherent characteristics of the target object can be obtained;

[0035] Based on the inherent characteristics of the target object and their performance under different interferences, continuous identification features are obtained;

[0036] Target objects are identified using continuous recognition features;

[0037] If recognition fails

[0038] The point cloud coordinates of continuously recognizable features in radar point cloud data are obtained by using multi-dimensional spatial backtracking;

[0039] Complete point cloud data is obtained by fine-tuning the point cloud coordinates of continuously identified features.

[0040] The method of obtaining the inherent characteristics of the target by transmitting radar point cloud data in different directions includes setting the direction of the target's movement as the X-axis direction, and then selecting the projection direction as the Y-axis or Z-axis direction.

[0041] The continuous identification features are obtained based on the inherent characteristics of the target object and its performance under different interferences. This includes manually observing and determining whether the projected point cloud has relatively clear identification features, whether the features are stable and continuous, whether they are easily lost due to interference from position or other factors, whether the time of loss is greater than 1 second, and whether the number of times they are lost is greater than 3.

[0042] The method of identifying target objects using continuous recognition features includes using an AI model to identify target objects based on continuous recognition features.

[0043] The point cloud coordinates of the continuous identification features of radar point cloud data obtained by using multi-dimensional space backtracking include backtracking the continuous identification features of the target point cloud in the previous time period, mapping the local data of the continuous identification features to the same multi-dimensional data space, and obtaining a time-based continuous target feature point cloud set.

[0044] The method of obtaining the point cloud coordinates of the continuous identification features from radar point cloud data through multi-dimensional spatial backtracking also includes calculating the velocity, acceleration, and direction of the target feature point cloud in the continuous target feature point cloud set, and then obtaining the position of the target feature point cloud. The method of obtaining the point cloud coordinates of the continuous identification features from radar point cloud data through multi-dimensional spatial backtracking also includes calibrating the target feature point cloud position with the radar point cloud data, and fine-tuning the target feature point cloud position to obtain a repaired target point cloud, thus completing the identification process.

[0045] Specifically, it includes the following steps:

[0046] Step 1: After acquiring the radar point cloud data of the target, map the point cloud data in different directions to find and discover some inherent (non-temporary, such as deformation point cloud data caused by loading) features of the target. If the vehicle is traveling in the X-axis direction, the projection direction is generally selected to be the Y or Z direction, because the lidar is usually installed on the side or directly above the vehicle.

[0047] By projecting the target object at an appropriate angle, suitable recognition features can be found, thus effectively avoiding the impact of interference signals on the recognition effect.

[0048] When interference signals are too strong, causing even these hidden identification features to fail, the target at some point in time will inevitably be lost. In this case, further target repair can be achieved through higher-dimensional projection effects. The process is as follows: By retrospectively analyzing the target point cloud feature data from a previous period, the local data of these features are mapped onto the same multi-dimensional data space. This forms a continuous time-based target feature point cloud set. The changing trends of these point clouds over time are analyzed, and the target's motion patterns, i.e., velocity and direction, are calculated. By predicting the target's velocity and direction, the positions of these features at a specific time scale are determined. The matching degree between the target features and the point cloud data is checked, and the feature target positions are fine-tuned, thereby repairing the target point cloud at this moment. The target point cloud repair work at a specific time point is completed.

[0049] Step 2: Observe the anti-interference performance of these features under different interferences to determine the features that perform well and can be continuously identified. This requires manual observation to determine whether the projected point cloud has relatively clear identifying features, whether these features are stable and persistent, whether they are easily lost due to location or other factors, whether the time of loss is greater than 1 second, and whether the number of times the features are lost is greater than 3.

[0050] Step 3: Identify the target object using the recognition features selected in Step 2. If identification fails, proceed to Step 4; otherwise, return the recognition result. Features are used for identification; identifying the features identifies the target object. Feature identification is performed using an AI model matching algorithm.

[0051] Step 4: Using multidimensional spatial backtracking technology, target point cloud data from a certain time period are imported into the same multidimensional data space at specific time intervals, retaining only valid point cloud data of the identified feature regions. Specifically, the point clouds of the identified feature regions are backtracked and fused over a period of time. The extent of fusion depends on the target's speed and radar scanning frequency. Generally, it is necessary to ensure that the target's travel distance covers the adjacent areas before and after the obscured area, as well as at least three frames of point cloud data beyond those adjacent areas, in order to statistically determine the target's travel speed.

[0052] Step 5: Analyze the cluster of identification feature points in the multidimensional data space. Based on the time and spatial coordinates of the identification features, calculate the target's trajectory, including direction, velocity, and acceleration. Calculate the target's direction, velocity, and acceleration by analyzing the locations of features in adjacent regions, thereby estimating its specific position at each moment it should appear in the occluded region.

[0053] Step 6: Calculate the spatial coordinates of the identified features on the specified time scale and return the coordinate data.

[0054] Step 7: Mark the position of the identified feature in the current point cloud, calculate its matching degree based on the geometric relationship between the target feature and other structures, and fine-tune its position based on the calculation result. Specifically, the calculated position is judged to match the geometric relationship of the identified feature on the target to determine if it conforms to the geometric relationship in the point cloud data. For example, a car door should appear one meter away from the front of the car, and the car door is the identified feature. If the front of the car is successfully identified in the radar point cloud, but the calculated position of the car door is greater than or less than one meter away from the front of the car, it proves that the calculated position of the identified feature is inaccurate and needs to be corrected using the actual point cloud data.

[0055] Step 8: Output the repaired complete point cloud data.

[0056] Example 2

[0057] This embodiment provides a system, including:

[0058] The data acquisition module is configured as follows:

[0059] A computer-readable storage medium storing a plurality of instructions adapted for loading and execution of the method by a processor of a terminal device.

[0060] A terminal device includes a processor and a computer-readable storage medium, the processor being configured to implement instructions; the computer-readable storage medium being configured to store a plurality of instructions adapted for loading by the processor and executing the method thereof.

[0061] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A point cloud data recognition method based on multi-dimensional feature target backtracking, characterized in that, include: Acquire radar point cloud data of the target object; By transmitting radar point cloud data in different directions, the inherent characteristics of the target object can be obtained; Based on the inherent characteristics of the target object and their performance under different interferences, continuous identification features are obtained; Target objects are identified using continuous recognition features; If recognition fails The point cloud coordinates of continuously recognizable features in radar point cloud data are obtained by using multi-dimensional spatial backtracking; Complete point cloud data is obtained by fine-tuning the point cloud coordinates of continuously identified features; The point cloud coordinates of the continuous identification features of radar point cloud data obtained by using multi-dimensional space backtracking include backtracking the continuous identification features of the target point cloud in the previous time period, mapping the local data of the continuous identification features to the same multi-dimensional data space, and obtaining a time-based continuous target feature point cloud set.

2. The point cloud data recognition method based on multi-dimensional feature target backtracking according to claim 1, characterized in that, The method of obtaining the inherent characteristics of the target by transmitting radar point cloud data in different directions includes setting the direction of the target's movement as the X-axis direction, and then selecting the projection direction as the Y-axis or Z-axis direction.

3. The point cloud data recognition method based on multi-dimensional feature target backtracking according to claim 2, characterized in that, The continuous identification features are obtained based on the inherent characteristics of the target object and its performance under different interferences. This includes manually observing and determining whether the projected point cloud has relatively clear identification features, whether the features are stable and continuous, whether they are easily lost due to interference from position or other factors, whether the time of loss is greater than 1 second, and whether the number of times they are lost is greater than 3.

4. The point cloud data recognition method based on multi-dimensional feature target backtracking according to claim 3, characterized in that, The method of identifying target objects using continuous recognition features includes using an AI model to identify target objects based on continuous recognition features.

5. The point cloud data recognition method based on multi-dimensional feature target backtracking according to claim 4, characterized in that, The method of obtaining the point cloud coordinates of the continuous identification features of radar point cloud data by using multi-dimensional spatial backtracking also includes calculating the velocity, acceleration and direction of the target feature point cloud in the continuous target feature point cloud set, and then obtaining the position of the target feature point cloud.

6. The point cloud data recognition method based on multi-dimensional feature target backtracking according to claim 5, characterized in that, The method of obtaining the point cloud coordinates of the continuous identification features of radar point cloud data by using multi-dimensional spatial backtracking also includes calibrating the target feature point cloud position and radar point cloud data, and fine-tuning the target feature point cloud position to obtain the repaired target point cloud and complete the identification.

7. A point cloud data recognition system based on multi-dimensional feature target backtracking, characterized in that, include: The data acquisition module is configured to acquire radar point cloud data of the target object; The identification module is configured to obtain the inherent features of the target object by transmitting radar point cloud data in different directions; obtain continuous identification features based on the performance of the inherent features of the target object under different interferences; and identify the target object using the continuous identification features. The calculation module is configured to, if recognition fails, use multi-dimensional space backtracking to obtain the point cloud coordinates of the continuously identified features of the radar point cloud data; and obtain complete point cloud data by fine-tuning the point cloud coordinates of the continuously identified features.

8. A computer-readable storage medium storing a plurality of instructions, characterized in that, The instructions are adapted to be loaded and executed by the processor of the terminal device as described in claim 1, which is a point cloud data recognition method based on multi-dimensional feature target backtracking.

9. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is configured to implement instructions; and the computer-readable storage medium is configured to store multiple instructions, characterized in that, The instructions are adapted to be loaded and executed by a processor as described in claim 1, which is a point cloud data recognition method based on multi-dimensional feature target backtracking.

Images