運用移動攝影機於移動影像特徵之偵測與追蹤

Detection and Tracking of Moving Image Features Using Moving Cameras

黃逸展(Yi-Jhan Huang)

指導教授 ： 王銀添

淡江大學/工學院/機械與機電工程學系碩士班/碩士(2013年)

https://doi.org/10.6846/TKU.2013.01108

摘要

本論文發展移動攝影機偵測與追蹤移動物體演算法，應用於機器人同時定位、建圖、與移動物體追蹤。本研究專注在移動特徵偵測與追蹤程序的規劃，總計探討與測試三種不同方法：稱為非靜態物件、移動物件、與廣義物件等偵測與追蹤方法。非靜態物件方法是文獻所提出的方法，利用本質矩陣的運算去區別非靜態與靜態的物件。但是，此方法限制影像中的特徵個數，使得部分移動特徵無法順利被偵測。本論文提出移動物件與廣義物件兩種方法進行移動特徵的偵測。移動物件方法放寬非靜態物件方法的特徵個數之限制，並且加入多重過濾程序，以便偵測所有的移動特徵。廣義物件方法則是估測所有影像特徵的狀態變數，所得到的狀態訊息可以提供做為靜態與移動物件的選擇。

關鍵字

移動物體偵測 ； 移動物體追蹤 ； 本質矩陣 ； 廣義物件

並列摘要

This thesis presents an algorithm of moving object detection and tracking using moving cameras. The developed algorithm is applied to robot simultaneous localization, mapping, and moving object tracking. The research focuses on the development of procedures for moving feature detection and tracking. In this thesis, three methods are investigated and tested, namely the methods of detecting non-static objects, moving objects, and generalized objects. Non-static object method is the procedure in the literature which distinguishes non-static image features from stationary features based on the essential matrix calculation. However, some moving features could not be detected due to the restriction on feature number in one image. The moving-object and generalized-object methods are proposed in this thesis. The moving-object method releases the restriction on feature number in an image and detects all possible moving objects by using multiple filtering processes. The generalized-object method estimates the state variables of all image features and provide the state information for the detection of moving and stationary features.

並列關鍵字

Moving Object Detection(MOD) ； Moving Object Tracking(MOT) ； Essential Matrix ； Generalized Objects

參考文獻

[3] 邱明璋，基於極線限制條件之單眼視覺式移動物體偵測與追蹤，淡江大學機械與機電工程學系碩士論文，2011。

[9] 陳國瑋，使用線上稀疏呈現的物件模型進行機器人並行追蹤與建圖，淡江大學機械與機電工程學系碩士論文，2012。

[16] 王琮鴻，多攝影機輔助機器人巡航，淡江大學機械與機電工程學系碩士論文，2013。

[1] T. Bailey, and H. Durrant-Whyte, 2006, Simultaneous Localization and Mapping (SLAM):

PartⅡ, IEEE Robotics and Automation, pp.108-117, September