戰場偵察長久以來一直都是戰術思維的重心，近年來有越來越多的偵察任務是利用機器來完成的，這些機器配有內建式感應器及自動目標辨識系統。 我們提出一個架構來實作自動目標辨識系統的偵測function利用梯度影像分析及直線偵測技術，將灰階影像中人造物體的概略輪廓描繪出來。首先，我們使用Sobel運算以取得影像的梯度，接著使用含有區域法則的模糊影像對比增強以去除背景及增強訊號弱的及訊號強的邊界；經過二元化、小區塊去除、及細線化後，我們使用改良式霍式轉換以偵測長的直線；利用這些直線，可以標示出可疑區域並利用這些區域以產生初始的物件輪廓；最後，我們利用適應式主動輪廓模型來進行輪廓趨近。 我們將之實作於一般的個人電腦上，而實驗結果顯示此架構能適用於大多數的環境條件下。 Reconnaissance has for centuries been at the heart of all thinking about infantry tactics. Nowadays, reconnaissance is increasingly assigned to machines. These machines are equipped with build-in sensors and automatic target recognition system (ATR) in it. We proposed a framework to perform the detecting phase in ATR systems. This system can label approximate man-made object contours in gray-level images via gradient image analysis and straight lines detection. We first use the Sobel operator to produce a gradient image. Then, use local fuzzy image contrast enhancement with a region criterion to degrade background and enhance both weak and strong edges. After the processes of binarization, small component removal, and edge thinning, we apply the modified Hough transform to detect long straight lines. Via these lines, we can label the region of interest and use them to produce initial object contours. At last of all, we apply the adaptive active contour model to perform contour approximation. Our experiment is performed on a PC and the experimental result shows that it works well under most environmental condition.
