基於三維光達之動態物件偵測與追蹤

Dynamic Object Detection and Tracking Based on 3D LiDAR

吳柏賢(Bo-Xian Wu)

指導教授 ： 李世安

淡江大學/工學院/電機工程學系人工智慧機器人碩士班/碩士(2022年)

https://doi.org/10.6846/TKU.2022.00605

摘要

本論文提出一種基於三維光達之動態物件偵測與追蹤之方法。本方法可根據三維光達所取得之點雲資訊進行資料處理後，可以獲得周圍環境之物件資訊。並且從物件資訊中分辨出動態物件與靜態物件後再進行動態物件位置的追蹤。本方法主要分成四個部分：(1)點雲分割、(2)點雲聚類、(3)匹配系統、以及(4)追蹤系統。在點雲分割部分，本論文使用三維光達作為感測器取得環境資訊，並將接收到之點雲資料分割成地面點雲、牆面點雲以及物件點雲三個部分，地面與牆面點雲會在後續資料處理上產生干擾，因此需要先行濾除。在點雲聚類部分，依據物件點雲之間的距離，將物件點雲區分為數個群集，並計算出每個群集之中心位置，獲得各個物件在空間中之分布狀況。在匹配系統部分，本論文將各個物件在不同時間下的分布位置進行匹配，記錄下物件在不同時間下的位置分布情況。在追蹤系統部分，依據匹配系統所紀錄之物件位置，推算出物件下一步的行徑位置，增強機器人對環境的感知能力。本論文分別以模擬器環境以及現實環境進行測試。在實驗結果上，實驗平台都能藉由三維光達感測器數據獲得周圍環境狀態，並實現動態物件追蹤之功能。

關鍵字

三維光達 ； Gazebo模擬器 ； 物件偵測 ； 物件追蹤 ； 距離感測

並列摘要

In this thesis, a dynamic object detection and tracking method based on 3D LiDAR is proposed for a robot sensing, so that robot can obtain the object information of the surrounding environment after data processing according to the point cloud information obtained by the 3D LiDAR. Then distinguish dynamic and static objects from object information, Track the position of the dynamic object and mark the path direction. There are four main parts:(1) point cloud segmentation, (2) point cloud clustering, (3) matching system and (4) tracking system. In point cloud segmentation, a 3D LiDAR is used to obtain the environmental information, and divides the received point clouds data into three parts: ground point clouds, wall point clouds and object point clouds. The ground and wall point clouds will be a huge distraction in data processing, so it needs to be filtered out first. In point cloud clustering, according to the distance between each object point clouds, the object point cloud is divided into several clusters, and the center position of each cluster is calculated to obtain the distribution of each object in space. In matching system, the position distribution of each object at different time is matched, then record the position of the object at different time according to the matching result. In tracking system, according to the position of the object recorded by the matching system, the next move position of each object is calculated to enhance the robot's ability to perceive the environment. In this thesis, the simulation environment and the real environment are used for testing. In the results, the robot can obtain the surrounding environment status through the data of the 3D LiDAR, and realize the function of dynamic object tracking.

並列關鍵字

3D LiDAR ； Gazebo simulator ； object detection ； object tracking ； range sensing

參考文獻

[1] TESLA Model 3，URL：https://www.tesla.com/zh_tw/model3，2022/07/04存取。

[2] Waymo，URL：https://waymo.com/intl/zh-tw/，2020/07/04存取。

[3] G. Tang, C. Tang, C. Claramunt, X. Hu, and P. Zhou,” Geometric A-star algorithm: an improved A-star algorithm for AGV path planning in a port environment,” IEEE Access, vol. 9, pp. 59196-59210, 2021.

[4] Amazon Robotics，URL：https://www.amazon.jobs/en/teams/amazon-robotics，2021/12/07存取

[5] B. Zhou, Y. He, K. Qian, X. Ma, and X. Li, ”S4-SLAM: A real-time 3D LIDAR SLAM system for ground/watersurface multi-scene outdoor applications.” Autonomous Robots, vol. 45, no. 1, pp. 77-98, 2021.