研究生: |
許哲維 Che-Wei Hsu |
---|---|
論文名稱: |
基於對比學習對未知目標進行動態追蹤與避障 Dynamically Target Tracking and Obstacle Avoidance with Contrastive Learning |
指導教授: |
施慶隆
Ching-Long Shih |
口試委員: |
施慶隆
Ching-Long Shih 王乃堅 Nai-Jian Wang 李文猶 Wen-Yo Lee 吳修明 Hsiu-Ming Wu |
學位類別: |
碩士 Master |
系所名稱: |
電資學院 - 電機工程系 Department of Electrical Engineering |
論文出版年: | 2022 |
畢業學年度: | 110 |
語文別: | 中文 |
論文頁數: | 89 |
中文關鍵詞: | 影像分割 、前景提取 、對比學習 、移動機器人之追蹤與避障 、懲罰式A*搜索演算法 |
外文關鍵詞: | Image segmentation, Foreground extraction, Contrastive learning, Target tracking and obstacle avoidance of mobile robot, Penalty A* search algorithm |
相關次數: | 點閱:278 下載:0 |
分享至: |
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報 |
本論文旨在運用色彩空間分群方法搭配多物件追蹤演算法實現即時的目標
追蹤移動機器人。首先由一般相機所提供的三通道彩色影像進行色彩空間的分群,
以此在任意單純背景下完成未知前景的影像分割。而後利用融合運動訊息與外觀
訊息的多物件演算法對沒有經過訓練的未知前景進行識別與匹配,再以透視投影
轉換的方式估算出已識別的前景於工作區內的具體位置。最後透過 A*搜索演算
法安排一條合適的路徑供移動機器人動態地追蹤目標物件且避開所有障礙物。本
文之特色為移動機器人的姿態估計完全依靠置於機器人頂部的二維標記 ArUco
完成,而不另外使用里程計、陀螺移與加速度計,同時也無須進行連續軌跡規劃,
僅由直線行進與原地旋轉完成追蹤及避障之雙重任務。另外,多物件追蹤演算法
的運動訊息由三階的卡爾曼濾波器完成估測,外觀資訊由對比學習的度量神經網
路完成,將兩者依場景獨立進行探討。在 A*搜索演算法中加入節點排序與懲罰
機制,以得到唯一的一條最佳路徑,因此可減少機器人旋轉次數與提升動態路徑
規劃的穩定性。
This paper aims to use the color space clustering method with multi-object tracking
algorithm to realize a real-time target tracking mobile robot. First, the three-channel
color image is provided by a camera, so as to complete the image segmentation of the
unknown foreground under any simple background. Then, the multi-object algorithm
of integrating motion information and appearance information is used to identify and
match the unknown foreground without training, and then the specific position of the
identified foreground in the work area is estimated by means of perspective projection
transformation. Finally, an appropriate path is planned by using the A* search algorithm
for the mobile robot to dynamically track the target object and avoid all obstacles. The
feature of this paper is that the position and orientation estimation of the mobile robot
is completely based on the two-dimensional marker ArUco which is placed on the top
of the robot, and without additional use of odometer, gyroscopic movement and
accelerometer. At the same time, there is no need for continuous trajectory planer, and
both tracking and obstacle avoidance are done only by straight-line translation and turnin-place rotation. In addition, the motion information of the multi-object tracking
algorithm is estimated by the third-order Kalman filter, and the appearance information
is completed by the metric neural network of contrastive learning, and these two are
independently developed according to the scene. A node sorting and penalty mechanism
is added to the A* search algorithm to obtain a unique optimal path, thus reducing the
motion of robot and improving the stability of the dynamic path planning.
1. Otsu, N., A Threshold Selection Method from Gray-Level Histograms. IEEE Transactions on Systems, Man, and Cybernetics, 1979. 9(1): p. 62-66.
2. Wang, D., et al., An Efficient Iterative Thresholding Method for Image Segmentation. 2017. 350: p. 657-667.
3. Upadhyay, P. and J.K. Chhabra, Kapur’s Entropy Based Optimal Multilevel Image Segmentation Using Crow Search Algorithm. Applied Soft Computing, 2020. 97: p. 105522.
4. Wojke, N., A. Bewley, and D.J.a.e.-p. Paulus Simple Online and Realtime Tracking with a Deep Association Metric. 2017. arXiv:1703.07402.
5. Bewley, A., et al. Simple Online and Realtime Tracking. 2016. arXiv:1602.00763.
6. Li, Q., et al. Kalman Filter and Its Application. in 2015 8th International Conference on Intelligent Networks and Intelligent Systems (ICINIS). 2015.
7. Kuhn, H.W., The Hungarian Method for the Assignment Problem. Naval Research Logistics Quarterly, 1955. 2(1-2): p. 83-97.
8. Zheng, L., et al., MARS: A Video Benchmark for Large-Scale Person Re-Identification. Vol. 9910. 2016. 868-884.
9. Zagoruyko, S. and N.J.a.e.-p. Komodakis Wide Residual Networks. 2016. arXiv:1605.07146.
10. Schroff, F., D. Kalenichenko, and J.J.a.e.-p. Philbin FaceNet: A Unified Embedding for Face Recognition and Clustering. 2015. arXiv:1503.03832.
11. Hermans, A., L. Beyer, and B.J.a.e.-p. Leibe In Defense of the Triplet Loss for Person Re-Identification. 2017. arXiv:1703.07737.
12. Song, H.O., et al. Deep Metric Learning via Lifted Structured Feature Embedding. 2015. arXiv:1511.06452.
13. Liu, W., Y. Li, and C. Koh. A Fast Maze-Free Routing Congestion Estimator with Hybrid Unilateral Monotonic Routing. in 2012 IEEE/ACM International Conference on Computer-Aided Design (ICCAD). 2012.
14. Schumann, A. and R. Stiefelhagen. Person Re-identification by Deep Learning Attribute-Complementary Information. in 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). 2017.
15. Chen, W., et al. Beyond Triplet Loss: A Deep Quadruplet Network for Person Re-Identification. 2017. arXiv:1704.01719.
16. Korf, R.E. and D.M. Chickering, Best-First Minimax Search. ICGA Journal, 1996. 19: p. 187-187.
17. Javaid, A., Understanding Dijkstra Algorithm. SSRN Electronic Journal, 2013.
18. Goto, T., T. Kosaka, and H. Noborio, On the Heuristics of A* or A Algorithm in ITS and Robot Path-Planning. 2003. 1159-1166 vol.2.