運用獨立成份分析於移動目標追蹤之研究

A Study on Tracking of Moving Targets Using Independent Component Analysis

周晏德(Yan-De Zhou)

指導教授 ： 陳冠宇

中原大學/工學院/機械工程研究所/碩士(2011年)

https://doi.org/10.6840/cycu201100649

摘要

偵測及追蹤移動目標物是智慧型機器視覺系統的重要功能之一，然而當背景與目標物同時改變的情況下，如何偵測移動目標物是其中的關鍵技術。獨立成份分析是一個統計和訊號處理的方法，可以在未知訊號源的線性混合資料中，將未知訊號分離成統計獨立的成份，進而擷出所欲求得之來源訊號。近年來，獨立成份分析已經成功應用在特徵萃取和影像分割上。為改善傳統影像相減法的缺點，本文提出運用獨立成份分析法，自複雜背景中擷取移動目標物，並且將之應用於具有二個影像追蹤平台的雙目標影像追蹤系統上，而做追蹤的運作，其中每個追蹤平台均具有二部攝影機及二個旋轉軸。其運作程序，先使用獨立成份分析法，自即時影像中分離出複雜背景與移動目標物，再經由必要的數位影像處理方法，估算出移動目標物在影像中的質心位置，最後轉換為移動目標物與攝影機之間的相對位置，使每個追蹤平台能夠追蹤各自的移動目標物，將之鎖定於攝影機的視野中心，以達到追蹤目標物之效果。實驗結果顯示本文提出的方案比傳統的影像相減法具有更好的效能。

關鍵字

機器視覺 ； 移動目標追蹤 ； 獨立成份分析

並列摘要

Detection and tracking of moving targets is one of the important functions of intelligent machine vision system. However, when the background and the target has changed at the same time, how to detect moving targets is the key technologies. Independent component analysis, which can separate the unknown source in the linear mixed data into statistically independent components, is a statistical and signal processing methods, and extract the desirous source signal. In recent years, ICA has been successfully applied in the feature extracted and the image separated. To improve the shortcomings of the traditional image subtraction, we propose the ICA method, which can capture the moving target from the complex background, to apply it on the target tracking system with two sets of platform of tracking dual-image, and then to perform the tracking works. There are two sets of CCD and its rotating pivots on each tracking platform. Its operational procedures is: Firstly, using analysis method of ICA separate the background and moving target from mixture and real-time image, and then by the necessary processing methods of digital image, to estimate the mass centroid position of the moving target in the image, and finally to transform into the relative position between the CCD and the moving target, so that let each tracking platform look for its own moving target respectively and to focus the moving target at the vision field central point of CCD for tracking target effect. Experimental results shows that this proposed scheme has better performance than the traditional image subtraction.

並列關鍵字

moving target tracking ； machine vision ； independent component analysis

參考文獻

[2] 陳衍廷，應用核心主成份分析及非平衡式SVM於影像中人臉偵測之研究，中原大學機械工程學系碩

[5] 張文龍，自主移動機器人之人機互動控制的設計與實現，中原大學機械工程學系碩士論文，2007。

[6] 呂育宏，自主移動機器人之智能導航控制系統的設計與發展，中原大學機械工程學系碩士論文，

[7] 蔡博仰，立體視覺伺服機器手臂之智慧追蹤控制的研究，中原大學機械工程學系碩士論文，2007。

[11]陳彥良，即時立體視覺物體追蹤系統，中原大學機械工程學系碩士論文，2003。