移體物體影像之即時追蹤系統

Abstract

本論文的目的在於利用固定背景下的連續影像，建立一個移動物影像的追蹤系統。首先分成移動物件的偵測與物件追蹤兩部份，在移動物偵測方面，運用影像處理的技術建立目前影像環境的背景影像，利用目前影像與背景影像的差異，雜訊的消除，以及臨界值的求取，將移動物件與背景分離，接著利用等位集合法(Level Set Method)將每一個物件的初始輪廓給分割出來。 而在物件追蹤方面，藉由等位集合法(Level Set Method)追蹤物件輪廓，以了解同一個移動物件在影像序列間的不同狀態，而物件輪廓的演化使用Narrow Band Method，由於Narrow Band Method限制了Level Set Method所需疊代計算的範圍，進而減少系統的運算量，降低追蹤移動物件所花費的時間。最後實驗結果顯示了我們的系統可以針對影像畫面上的多目標移動物件來追蹤，並且能達到即時性的要求。

The purpose of this thesis lies in using image sequences in the static background tobuild a image tracking system for moving objects. First of all, we separate the tracking system into two part－detecting and tracking. On the aspect of detecting moving object, we applied the technique of image processing to build the background image based on the present image environment. By the application of pixelwise difference between background image and current image, elimination of noise and the calculation of threshold value, we separated background and moving object,and then we extracted theorigional outline of each moving object by the applying "Level Set Method". As for the tracking, we tracked down the outline of moving objects with Level Set Method so that we could understand the status of the same moving objects within the image sequences. For the evolution of moving object outline we applied Narrow Band Method. Due to Narrow Band Method limited the range of iteration calculation that Level Set Method required, it therefore reduced the operation capacity of the system, and reduced the time spent on tracking moving objects. The outcome of the final experiment showed that our system could do tracking by aiming at multi-moving objects on the screen and fulfilled the instantaneous requirements.