Fire Tracking in Video Sequences Using Geometric Active Contours Controlled by Artificial Neural Network

Automatic fire and smoke detection is an important task to discover forest wildfires earlier. Tracking of smoke and fire in video sequences can provide helpful regional measures to evaluate precisely damages caused by fires. In security and surveillance applications, real-time video segmentation of both fire and smoke regions represents a crucial operation to avoid disaster. In this work, we propose a robust tracking method for fire regions using an artificial neural network (ANN) based approach combined with a hybrid geometric active contour (GAC) model based on Bayes error energy functional for forest wildfire videos. Firstly, an estimation function is built with local and global information collected from three color spaces (RGB, HIS and YCbCr) using Fisher's Linear Discriminant analysis (FLDA) and a trained ANN in order to get a preliminary fire pixel classification in each frame. This function is used to compute initial curves and the level set evolution parameters to control the active contour model providing a refined fire segmentation in each processed frame. The experimental results of the proposed tracking scheme proves its precision and robustness when tested on different varieties of scenarios whether wildfire-smoke video or outdoor fire sequences.