Sistema de detecção de fogo e detecção e seguimento de objectos

Abstract

This dissertation presents a system for early fire detection and object tracking in live video sequences obtained from fixed outdoor surveillance cameras. Focus is given to the challenges related to the actual deployment of the vision system. Namely, background subtraction, which is a key to determine which regions of the environment belong to the foreground, is performed in a windowed way for improved accuracy. To reduce the computational cost, an attentive mechanism is employed to focus a computationally expensive frequency analysis of potential fire regions. To promptly discriminate fire regions from fire colored moving objects, a new colour-based model of fire’s appearance and a new Wavelet-based model of fire’s frequency signature are proposed. Besides that, to reduce the false alarms on fire detection of moving object with fire-colored appearance, an innovative solution to integrate the results of the two algorithms is proposed. Namely, the movement of the tracked object on the environment is analyzed. In addition, camera-world mapping is approximated according to a GPS-based learning calibration process to generate geo-located alarms, and to estimate the object height in the image plane. Experimental results demonstrate the ability of the proposed model to robustly detect fires and track moving objects, even in the presence of severe occlusions. Concretely, an average success rate of 92.7 % to detect fire and 92.8 % to tracking objects at a processing frequency of 10 Hz shows the applicability of the model to real-life applications.