影像基礎的數位自動監控系統

Abstract

商業用的閉路電視監控系統多仰仗人工的監控與判斷，其本身只具有傳輸影像和記憶的功能，無法判斷所監控的地方是否有狀況發生，而記憶的功能也由於無法判斷狀況的發生而浪費於在不需要儲存記憶的地方。在現今的許多文獻之中，有許多方式來增進其自動判斷的功能，但大部分的方法皆需要大量的計算，導致無法線上工作。本文採用並改進了「點狀相關圖分析法」這種不需要大量計算，而且對於雜訊有相當高的強健性的方式，利用兩張影像的灰階值和其位置的相對關係重新排列出一組新的資料作為分析之用。藉由這一組新的資料可以判斷場景是否改變，且此方法對於雜訊也不需進行前處理的動作，而對於光度微量的變化也可以接受。經過實驗的結果證明，這個方式可以實際應用於線上的及時辨識及監控工作。

Many commercial CCTV security systems still need human operator for monitoring. These systems can automatically transmit and save images captured by camera. However, they can not distinguish whether or not suspicious activity is taking place in the field of view of camera. Because they can not detect suspicious activity automatically, most storage spaces are wasted on saving meaningless images. Many methods to automatically detect suspicious activity were studied recently. However, most of them are computationally too intensive to be useful in real-time system. In this paper, a modified scatter plot analysis is applied to detect variations in two consecutive picture frames. As a result, we can capture any activities taking place in the field of view of the camera. Since this method is very robust to noise, it can also tolerate variations in the luminance slightly. In addition, the experiment at result showed that this method is fast enough to be useful in a real-time system in practice.