結合雷射測距與單眼影像於嵌入式移動機器人之人體偵測與追蹤

Human Detection and Tracking for Embedded Mobile Robots by Integrating Laser-range-finding and Monocular Imaging

王偉航(Wei-Hang Wang)

指導教授 ： 陳永昌

國立清華大學/電機資訊學院/電機工程學系/碩士(2012年)

摘要

現代的服務型機器人有個重要的基礎課題,那就是人機互動(Human-robot Interaction)。而要達成這項任務,機器人必須具備有能偵測與跟蹤周圍的人的人 的能力。由其是對於自主移動機器人,能穩健地追蹤目標的能力是他們不可或缺的。因此,一個穩健的人體偵測與追蹤系統在機器人學中是一塊很重要的研究領域。

在本篇論文中,我們提出了一套裝載於移動平台的系統,其上搭載有雷射測 距儀與單眼攝影機。機器人能取得並結合單眼攝影機的影像以及雷射測距資訊, 有效率地偵測並追蹤周圍的人。我們使用一個由 Cascade Adaboost 架構訓練一組 人腳的雷射點幾何特徵而成的偵測器來偵測人腳。我們也修改了 C4 人體偵測器, 加入了雷射資訊並提出一個 Range C4 人體偵測器,它可以達到比 C4 偵測器更低的 false positive rate.

另外,這些被偵測到的人腳跟人體也進一步處理與融合以達成持續的跟蹤的目標,我們使用 global nearest neighbor(GNN)來做 data association,以及序慣式卡爾曼濾波器(sequential Kalman filter)和定速模型做為追蹤的策略。所以偵測到的資料與所以現有的追蹤目標之間的資料分配與聯結由 GNN 來決定, GNN 可以找出一個整體相似度合為最高的分配結果。而每個目標的狀態則是根據分配到的偵測結果依序用卡爾曼濾波器更新。我們也讓機器人在一般室內場所做了即時追踨的實驗來證明我們提出的方法的穩健與效率。

關鍵字

人體偵測 ； 腳偵測 ； 追蹤 ； 單眼相機 ； 雷射測距儀 ； 機器人

並列摘要

A fundamental issue for modern service robots is human–robot interaction. In order to perform such a task, these robots need to detect and track people in the surroundings. Especially, to track targets robustly is a indispensable capability of autonomous mobile robots. Thus , a robust human detection and tracking system is an important research area in robotics.

In this thesis, we present a system which is able to detect and track people efficiently by integrating laser measurements and monocular camera images information on mobile platform. A laser-based leg detector is used to detect human legs, which is trained by cascaded Adaboost with a set of geometrical features of scan segments. A visual human detector Range C4 is also proposed, which is modified from C4 human detector by adding laser range information. It achieves lower false positive rate than original C4 detector.

The detected legs or persons are fused and tracked by a global nearest neighbor (GNN) data association and a sequential Kalman filtering with constant velocity model strategies. Measurements are assigned to tracks by GNN which assigns measurement by maximum similarity sum, and track states are updated by using corresponded measurements sequen- tially. Several experiments are done and to demonstrate the robustness and efficiency of our system.

並列關鍵字

human detection ； leg detection ； tracking ； monocular camera ； laser-range-finder ； robot

參考文獻

[1] B. Schiele, M. Andriluka, N. Majer, S. Roth, and C. Wojek, “Visual people detection: Different models, comparison and discussion,” Proceedings of the IEEE International Conference on Robotics and Automation (ICRA), 2009.

[2] W. Zajdel, Z. Zivkovic, and B. J. A. Kro ̈se, “Keeping track of humans: Have I seen this person before?” Proceedings of the IEEE International Conference on Robotics and Automation (ICRA), pp. 2081–2086, April 2005.

[3] A. Ess, B. Leibe, K. Schindler, and L. van Gool, “Robust multiperson tracking from a mobile platform,” IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), vol. 31, no. 10, pp. 1831–1845, October 2009.

[4] D. Gavrila and S. Munder, “Multi-cue pedestrian detection and tracking from a moving vehicle,” International Journal of Computer Vision (IJCV), vol. 73, no. 1, pp. 41–59, 2007.

[6] F. D. la Torre, C. Vallespi, P. E. Rybski, M. Veloso, and T. Kanade, “Learning to track multiple people in omnidirectional video,” Proceedings of the IEEE International Conference on Robotics and Automation (ICRA), pp. 4150–4155, April 2005.