基於單眼視覺單一影像運動控制的自主式移動機器人之目標追蹤

Monocular Vision Single Image Based Motion Control for Autonomous Mobile Robot Target Tracking

黃柏勛(Bo-Shiun Huang)

指導教授 ： 周佑誠

中原大學/工學院/機械工程研究所/碩士(2012年)

https://doi.org/10.6840/cycu201200848

摘要

近年來自主式移動機器人的發展迅速，且被大量地應用在醫療照護、保全等各種領域。視覺追蹤系統之研究發開在擴展及強化自主式移動機器人之功能與應用層面，扮演一個關鍵角色。一個最佳的，或至少合適的視覺追蹤系統必須同時具備高準確度及使用少量軟硬體資源等特性。
本論文提出一個新的自主式移動機器人之即時目標追蹤法。首先，此方法利用粒子濾波技術預測移動目標在影像中之位置；由於粒子濾波的特性，所以此方法能有效地掌握線性與非線性之運動行為。另外，此方法使用簡單的數學運算將移動目標之影像資訊轉換為其真實座標資訊，所以此方法消耗少量的運算資源。最後，此方法採用單眼視覺法，亦即使用單一攝影機，可以使用較少的硬體資源來實現。
此方法會先針對圖像內目標物下一時刻的位置及大小進行估計，再由預測所得之資訊對應至物體真實位置，並控制移動機器人使目標物保持在攝影機之視野中央。本論文分別進行了線性運作與非線性動作的追蹤測試，並與卡爾曼濾波法進行比較。由實驗結果顯示，本文所提之方法皆能在以上實驗裡達到良好的追蹤效果，並且整體追蹤表現優於卡爾曼濾波法。

關鍵字

粒子濾波 ； 單眼視覺法 ； 目標追蹤 ； 移動機器人

並列摘要

Due to the rapid improvement of the autonomous mobile robot technology in recent years, autonomous mobile robots have been widely applied to a variety of domains such as medical operations, healthcare, and security. The development of visual tracking systems plays a key role in expanding and enhancing the functions and applications of autonomous mobile robots. An optimal, or at least suitable, visual tracking system should possess high accuracy and use few resources in hardware and software.
This thesis proposes a new motion control method, based on monocular vision and single image, for autonomous mobile robot target tracking. The proposed method predicts a moving target’s position in an image through a particle filter. Due to the stochastic properties of particle filtering, the proposed method can effectively and accurately handle both linear and nonlinear dynamic motions. In addition, the proposed method uses simple polynomial calculations to map a target’s virtual position to its real-world coordinates. Thus, the proposed method needs few software resources for computation. Moreover, the proposed method adopts the monocular vision approach, i.e., it uses a single camera, and therefore it needs few hardware resources for implementation.
The proposed method predicts a moving target’s position in an image, and calculates the virtual position’s real-world coordinates relative to a mobile robot. Based on to the target’s relative coordinates, the mobile robot is commanded to move towards the target in order to keep the target at the camera’s central field of view. Experimental results show that the proposed method can produce acceptable to good results in linear and nonlinear tracking experiments, and has an overall better tracking performance than the Kalman filter approach.

並列關鍵字

Particle filtering ； Monocular vision method ； Mobile robots ； Object tracking

參考文獻

2. Arulampalam, M.S., et al., A Tutorial on Particle Filters for Online Nonlinear/Non-Gaussian Bayesian Tracking. IEEE Transactions on Signal Processing, 2002. 50(2): p. 174-188.

3. Doucet, A., N.d. Freitas, and N. Gordon, eds. Sequential Monte Carlo Methods in Practice. 2001, Springer.

6. DeSouza, G.N. and A.C. Kak, Vision for mobile robot navigation: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2002. 24(2): p. 237-267.

8. Rahman, A., et al., An Image Based Approach to Compute Object Distance. International Journal of Computational Intelligence Systems, 2008. 1(4): p. 304-312.

11. Ying, W. and T.S. Huang, Nonstationary color tracking for vivion-based human-computer interaction. IEEE Transactions on Neural Networks, 2002. 13(4): p. 948-960.