Method and system for mapping and positioning in bulk material scene

Abstract

The invention relates to a method and a system for mapping and positioning in a bulk material scene, and belongs to the technical field of map construction. The method mainly comprises the following steps: collecting real-time point cloud data of a bulk cargo scene, identifying the point cloud data of a dynamic target through a trained PointRCNN dynamic target detection model, and obtaining a bounding box of the dynamic target; filtering the point cloud data of the dynamic target by adopting a BBOX filter; the filtered point cloud data are input into FAST-LIO for map construction; and carrying out target positioning based on the constructed map. According to the method, the dynamic target is accurately detected and removed, so that the mapping and positioning precision and robustness are improved, and important technical support and improvement are provided for application of intelligent bulk cargo processing and autonomous navigation systems.

Description

Technical Field

[0001] The present invention belongs to the technical field of map construction, and relates to a method and system for mapping and positioning in a bulk material scene by using PointRCNN to improve FAST-LIO. Background Art

[0002] In bulk material scenarios, FAST-LIO can be used for mapping and positioning, providing accurate environmental information, which can significantly improve navigation and operation accuracy, thereby improving work efficiency and safety. However, dynamic targets (such as pedestrians and vehicles) will have a significant impact on map construction, causing the sensor to generate noisy point cloud data, which will lead to a series of problems: First, the positioning drift problem. The interference of dynamic targets may cause the algorithm to produce incorrect feature matching, which in turn affects the accuracy of pose estimation, causing positioning drift, and even causing the entire mapping process to fail; second, the map distortion problem. The movement of dynamic objects will generate unnecessary noise on the static map, resulting in the generated map being inaccurate and difficult to reflect the real environment.

[0003] Therefore, there is an urgent need for a method to remove dynamic targets in point cloud data, overcome the shortcomings of the FAST-LIO mapping and positioning algorithm, and improve the accuracy and robustness of mapping and positioning in complex bulk material scenes. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method and system for mapping and positioning in bulk material scenes, use PointRCNN to improve the mapping of FAST-LIO in bulk material scenes, solve the problem of dynamic target interference in bulk material scenes, and improve the performance and reliability of the system in mapping and positioning tasks.

[0005] To achieve the above-mentioned objectives, the present invention provides, on the one hand, a method for mapping and positioning in a bulk material scene, which comprises: collecting real-time point cloud data of the bulk material scene, identifying dynamic targets through a trained dynamic target detection model based on PointRCNN, and obtaining a bounding box (BBOX) of the dynamic target; using a BBOX filter to filter out the point cloud data of the dynamic target; inputting the filtered point cloud data into FAST-LIO for map construction; and positioning the target based on the constructed map.

[0006] Furthermore, the training process of the Point RCNN dynamic target detection model includes: collecting dynamic target point cloud data of bulk material scenes, making a data set and dividing it into a training set and a test set; configuring the PointRCNN dynamic target detection model; training and optimizing the model through the training set; after reaching the predetermined number of training times, using the test set to evaluate whether the model meets the accuracy requirements, if it meets the accuracy requirements, output the model, otherwise retrain the model.

[0007] Furthermore, target positioning based on the constructed map includes: real-time collection of point cloud data of bulk material scenes; using the trained PointRCNN dynamic target detection model to identify dynamic targets and obtain bounding boxes corresponding to dynamic targets; filtering out point cloud data of dynamic targets through a BBOX filter; matching the filtered point cloud data of bulk material scenes with the constructed map to obtain the current positioning of the target.

[0008] On the other hand, the present invention provides a system for mapping and positioning in a bulk material scene, the system comprising:

[0009] Point cloud data acquisition module, which collects real-time point cloud data in bulk material scenes through lidar;

[0010] A dynamic target detection module is used to identify dynamic targets in bulk material scenes and obtain bounding boxes corresponding to the dynamic targets;

[0011] A map construction module receives data from the point cloud data acquisition module and the dynamic target detection module, filters out the dynamic target point cloud data using a BBOX filter, and constructs a map based on the filtered point cloud data; and

[0012] The positioning module receives data from the point cloud data acquisition module, the dynamic target detection module and the map construction module to locate the target. Specifically, it receives data from the point cloud data acquisition module and the dynamic target detection module, and uses the BBOX filter to filter out the dynamic target from the point cloud data collected by the point cloud data acquisition module, and then uses the filtered point cloud data to match the map constructed by the map construction module to obtain the current positioning result of the target.

[0013] The beneficial effects of the present invention are:

[0014] (1) The present invention uses PointRCNN to detect dynamic targets. By effectively removing dynamic targets, the possibility of mismatching and positioning drift is reduced, the stability and accuracy of pose estimation are improved, and the entire mapping process is made more reliable and accurate.

[0015] (2) In the map construction stage, by removing the noise and interference introduced by dynamic targets, the generated map is clearer and more realistic, which can better reflect the actual situation of the bulk material scene and improve the accuracy and usability of the map;

[0016] (3) The present invention combines PointRCNN dynamic target detection technology and FAST-LIO mapping and positioning algorithm, which can effectively process dynamic targets, so that the system can show stronger adaptability in changeable bulk material scenarios, effectively respond to environmental changes, maintain good performance, and improve the robustness and reliability of the entire system.

[0017] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0019] Figure 1 It is a schematic diagram of the process of the method of the present invention;

[0020] Figure 2 The structure diagram of the system of the present invention is shown in FIG. DETAILED DESCRIPTION

[0021] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0022] In order to solve the problem of being disturbed by dynamic targets when using FAST-LIO to build maps in complex bulk material scenes, an embodiment of the present invention proposes a method for improving FAST-LIO mapping using the PointRCNN algorithm, thereby improving the robustness and accuracy of mapping and positioning. The method mainly includes: collecting point cloud data of bulk material scenes; training the PointRCNN dynamic target detection model; collecting mapping data in dynamic scenes; in the FAST-LIO mapping stage, according to the dynamic target information detected by PointRCNN, performing dynamic target removal processing on it, and the processed point cloud is used for inter-frame matching and map construction; in the positioning stage, PointRCNN is used to detect dynamic targets, and the dynamic targets are removed from the point cloud, and the removed point cloud is used for positioning.

[0023] like Figure 1 As shown, the specific contents of this method are as follows:

[0024] The first is the process of building the map, including:

[0025] 1. Collect point cloud data of bulk material scenes and create data sets.

[0026] 2. Train the PointRCNN dynamic target detection model;

[0027] Specifically, the training process includes dividing the prepared data set into a training set and a test set; configuring a dynamic target detection model; using the training set to iteratively train and optimize the model; after reaching a predetermined number of training times, evaluating whether the model meets the accuracy requirements, if so, outputting the model, otherwise re-training to adjust the optimization parameters.

[0028] 3. Dynamic scene mapping data collection: specifically, using a data collection vehicle to collect mapping data during normal operations.

[0029] 4. Point cloud dynamic target removal processing;

[0030] Specifically, the trained PointRCNN dynamic target detection model is used to identify dynamic targets, obtain the BBOX corresponding to dynamic targets such as pedestrians and vehicles, and then filter out the corresponding point cloud data through the BBOX filter to obtain the point cloud data used for matching and mapping.

[0031] 5. Input the point cloud data without dynamic targets into the input data of the FAST-LIO algorithm to obtain the final mapping result.

[0032] The second is the positioning process, including:

[0033] 6. Obtain real-time point cloud data in dynamic scenes through laser radar; laser radar can be mounted on engineering vehicles to achieve positioning of engineering vehicles;

[0034] 7. Use the trained PointRCNN dynamic target detection model to identify dynamic targets, such as pedestrians and vehicles, obtain the BBOX corresponding to the dynamic target, and then filter out the corresponding dynamic target point cloud through the BBOX filter to obtain the point cloud data for positioning.

[0035] 8. Match the point cloud data after filtering out dynamic targets with the map data to obtain the current positioning result of the target.

[0036] In another embodiment of the present invention, a bulk material scene map construction and positioning system is provided, which uses PointRCNN to improve the map construction of the FAST-LIO algorithm, such as Figure 2 As shown, the system includes:

[0037] Point cloud data acquisition module, which collects real-time point cloud data in bulk material scenes through LiDAR;

[0038] The dynamic target detection module uses the PointRCNN model to identify dynamic targets and obtain the BBOX corresponding to the dynamic targets;

[0039] A map construction module receives data from the point cloud data acquisition module and the dynamic target detection module, and filters out dynamic targets from the point cloud data collected by the point cloud data acquisition module based on the BBOX filter, and then uses the FAST-LIO algorithm to construct a map based on the point cloud data after filtering out the dynamic targets;

[0040] The positioning module receives data from the point cloud data acquisition module and the dynamic target detection module, and filters out dynamic targets from the point cloud data collected by the point cloud data acquisition module based on the BBOX filter, and then matches the point cloud data after filtering out the dynamic targets with the map data stored in the map construction module to obtain the current positioning result of the target.

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for mapping and positioning in a bulk material scene, characterized by: The method includes: collecting real-time point cloud data of bulk material scenes, identifying dynamic targets through a trained dynamic target detection model, and obtaining a bounding box of the dynamic target; using a BBOX filter to filter out the point cloud data of the dynamic target; inputting the filtered point cloud data into FAST-LIO for map construction; and locating the target based on the constructed map.

2. The method according to claim 1, characterized in that: The dynamic target detection model is a PointRCNN-based model.

3. The method according to claim 1 or 2, characterized in that: The training process of the dynamic target detection model based on Point RCNN includes: collecting dynamic target point cloud data of bulk material scenes, making a data set and dividing it into a training set and a test set; configuring the PointRCNN dynamic target detection model; training and optimizing the model through the training set; after reaching a predetermined number of training times, using the test set to evaluate whether the model meets the accuracy requirements, if the accuracy requirements are met, the model is output, otherwise the model is retrained.

4. The method according to claim 1, characterized in that: The target positioning based on the constructed map includes: collecting point cloud data of the bulk material scene in real time; using a trained dynamic target detection model to identify dynamic targets and obtain a bounding box corresponding to the dynamic target; filtering out the point cloud data of the dynamic target through a BBOX filter; matching the filtered point cloud data of the bulk material scene with the constructed map to obtain the current positioning of the target.

5. A system for mapping and positioning in bulk material scenes, characterized by: The system includes: Point cloud data acquisition module, used to collect real-time point cloud data in bulk material scenes; A dynamic target detection module is used to identify dynamic targets in bulk material scenes and obtain bounding boxes corresponding to the dynamic targets; a map construction module, receiving data from the point cloud data acquisition module and the dynamic target detection module, filtering out the dynamic target point cloud data using a BBOX filter, and constructing a map based on the filtered point cloud data; and The positioning module receives data from the point cloud data acquisition module, the dynamic target detection module and the map construction module to locate the target.

6. The system according to claim 5, characterized in that: The method for implementing positioning in the positioning module includes: receiving data from the point cloud data acquisition module and the dynamic target detection module, and using the BBOX filter to filter out the dynamic target from the point cloud data collected by the point cloud data acquisition module, and then using the filtered point cloud data to match the map constructed by the map construction module to obtain the current positioning result of the target.

7. The system according to claim 5, characterized in that: The dynamic target detection module performs dynamic target detection through a dynamic target detection model based on PointRCNN.

8. The system according to claim 7, characterized in that: The training process of the dynamic target detection model based on PointRCNN includes: collecting dynamic target point cloud data of bulk material scene through a point cloud data acquisition module, making a data set and dividing it into a training set and a test set; configuring the PointRCNN dynamic target detection model; training and optimizing the model through the training set; after reaching a predetermined number of training times, using the test set to evaluate whether the model meets the accuracy requirements, if the accuracy requirements are met, the model is output, otherwise the model is retrained.

9. The system according to claim 5, characterized in that: The map construction module uses FAST-LIO to construct the map.

10. The system according to claim 5, characterized in that: The point cloud data acquisition module acquires point cloud data through a laser radar.

Images