本文主要目的為設計出一套追蹤系統，使移動機器人能在多目標物同時移動之情況下，跟蹤特定目標物，並同時閃避障礙物。本文著重討論人體追蹤；先使用人體偵測系統，檢測出畫面中所有的人體訊息。接著，藉由檢測每一個人體的外觀資訓及位置資訊，來判別並鎖定想追蹤的目標物。本文的人體追蹤系統主要參考物件跟蹤的經典演算法deepsort。經過重新設計架構內容後，使跟蹤系統保有原先的跟蹤能力。同時，也改善原架構在追蹤目標物脫離畫面後，便再無法重新追蹤其的致命缺點。在偵測人體的部分，本文的設計雖然與原系統一樣，採用預先訓練好的模型，但在特徵提取的部分，則無須和原架構一樣，必須額外訓練一個提取特徵的模型。最後將RGB 影像結合深度影像，進一步得知目標物及障礙物的世界座標後，最後進行路徑規劃，達成跟蹤目標物及避開障礙物的任務。

The main purpose of this paper is to design a tracking system, so that the mobile robot can track a specific target when multiple targets are moving. When tracking the specific target, the mobile robot also has to dodge the obstacles in front of it at the same time. The paper focuses on the object tracking. First, we use the human body detection system to detect the entire human body that appears on the screen. Then, by detecting the appearance information and motion information of each human body, the target to be tracked is identified and locked. The human body tracking system in this paper mainly refers to the classic algorithm deepsort for object detection. After redesigning the concept of deepsort, the tracking system cames up with not only retains its original tracking system ability but also improves its fatal flaw too, which is that it cannot track the specific target again after the target leaves its sight. In the problem of detecting the human body, although the same pre-trained model is used as the original tracking system does. When running the feature extraction stage, it does not need to train an additional feature extraction model like the original system has to. After the RGB image is combined with the depth image, tracking the world coordinates of the specific target and obstacles are obtained. Finally, the tracking path is planned for the mobile robot to follow the specific target and avoid the obstacles in front of it.

1. Nicolai Wojke, Alex Bewley, Dietrich Paulus, Simple Online and Realtime Tracking with a Deep Association Metric, 2017.
2. Esmael Hamuda, Improved Image Processing-based Crop Detection Using Kalman Filtering and the Hungarian Algorithm, 2018.
3. T. Hellstrom, Kinematics Equations for Differential Drive and Articulated Steering, Department of Computing Science, Umea University, 2011.
4. Redhwan Algabri and Mun-Taek Choi, Deep-Learning-Based Indoor Human Following of Mobile Robot Using Color Feature, 2020.
5. Songmin Jia, Liang Zhao, Xiuzhi Li, Wei Cui, Jinbo Sheng, Autonomous Robot Human Detecting and Tracking Based on Stereo Vision, 2011.
6. Ma'te' Petho˝, A'da'm Nagy, Tama's Zsedrovits, A Bio-motivated Vision System and Artificial Neural Metwork for Autonomous UAV Obstacle Avoidance, 2020.
7. Piyush Juyal and Sachin Sharma, Locating People in Real-World for Assisting Crowd Behaviour Analysis Using SSD and Deep SORT Algorithm, 2021.
8. 施慶隆, 李文猶, 機電整合控制：多軸運動設計與應用, 全華圖書, 2015.