透過您的圖書館登入
IP:3.146.176.81
  • 學位論文

利用色彩資訊及深度資訊之室內行動機器人對行人偵測與追蹤

Pedestrian Detection and Tracking with Indoor Mobile Robot Using both Color Information and Distance Information

指導教授 : 連豊力
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


生活之中,人機互動是一門重要的學問。如果機器人可以隨時隨地去跟人互動,那麼在生活上必定會便利不少。其中與人互動的情況中,追蹤行人就是其中一個必須研究的課題。機器人追蹤行人的問題,主要分成兩類:自我定位與地圖建構、行人偵測與追蹤。這些問題都是在做機器人追蹤行人時會碰到的項目。 以自我定位與地圖建構來說,要在未知路徑的情況下,得到行動機器人周邊及本身存在的資訊是一件困難的事。在本篇論文裡,主要解決行動機器人編碼器的硬體因素造成殘存誤差。利用粒子群最佳化演算法去找出更準確的機器人位置,然後利用這些資訊可以估測出機器人周遭環境地圖。由於雷射測距儀是一個相當精確的儀器,所以得到的距離資訊準確度是很高,在實驗部分會有驗證。在得到行動機器人的位置和周遭地圖之後,就可以分辨出靜態物地圖和動態物地圖。對行動機器人而言,是非常有用的資訊。然而,動態物的地圖推測會有一些問題,這部分會在論文中詳細的描述到。 另外,在行人偵測部分,除了用雷射測距儀所得到的動態點之外,也可以用點的分群、大小以及資料可信度去判斷。另外,色彩資訊的添加可以當成是一個更強韌的條件。偵測邊緣形狀的霍式圓形轉換、色彩判定以及頭的所在位置都是偵測行人的準則。利用這些準則,幾乎可以更精確的去判斷行人。基於行動機器人追蹤,由於追蹤目標行人在生活環境中可能會有靜態障礙物擋住或者附近突然出現的行人,造成雷射測距儀的資料誤判。場景設定是在一般研究室或宿舍,通常人的穿著色彩分佈及紋理分佈會不盡相同,所以可以透過色彩資訊去加以判斷。並且使用距離資訊空間連續性去做一個強韌性判斷的準則。利用這些準則,就可以強韌地去判斷行人,並且去做目標行人的追蹤。在本篇論文中,最主要的就是用色彩資訊去解決在使用雷射測距儀時,行人偵測的判斷以及突然出現的人進而造成追蹤判斷的錯誤。 在實驗結果顯示出在這些環境當中,可以解決雷射測距儀資料連結的不足,並且可以做到追蹤目標行人。

並列摘要


In daily life, that mobile robot communicates with pedestrian needs many tasks. The applications are used in guided vehicle, shopping cart, or office assistance. In this thesis, the tasks include self-localization, mapping, pedestrian detection, and target pedestrian tracking in unknown indoor environment. To do self-localization and mapping, the accurate odometry of mobile robot is important. However, skidding and slipping can induce that the odometry is not equal to the real distance. In this thesis, particle swarm optimization (PSO) algorithm correct odometry in unknown indoor environment. The combination of the self-localization and the mapping is referred to as the simultaneous localization and mapping (SLAM) [39: Birk & Carpin 2006]. In the SLAM, once odometry of mobile robot is known, building a map is also a task which can be effectively solved at the same time [39: Birk & Carpin 2006]. Afterward, moving object detection is based on the precise map. After the moving objects are detected, the next steps are pedestrian detection and target pedestrian tracking. In the pedestrian detection, the color image is regarded as an additional condition for the judgment based on the laser range finder (LRF) scan. In target pedestrian tracking, owing to pillars hindering or new pedestrians appearing, the data association may be error between two consecutive LRF scans. In this thesis, a method based on color distribution and color texture to track pedestrian in color images is proposed. The experiment demonstrates the target pedestrian in the new pedestrian appearing and the pillars hindering. Through the experiments, the performance of pedestrian detection and target pedestrian tracking is not good. However, the performance of pedestrian in color image is low owing to the resolution. In the future, the detection and tracking moving object (DATMO) with LRF scan in Chapter 3 can combine the pedestrian detection and target pedestrian tracking with color image in this thesis.

參考文獻


[1: Seifert & Kay 1995]
R. W. Seifert and M.G. Kay, “Evaluation of AGV routing strategies using hierarchical simulation,” in Proceedings of Winter Simulation Conference, Arlington, USA, pp. 850-856, Dec. 3-6, 1995.
[2: Nishimura et al. 2007]
S. Nishimura, H. Takemura, and H. Mizoguchi, “Development of attachable modules for robotizing daily items -person following shoppingcart robot-,” in Proceedings of IEEE Robotics and Biomimetics, Sanya, China, pp. 1506-1511, Dec. 15-18, 2007.
[3: Chen et al. 2011]

延伸閱讀