Neuronal mapped hybrid background segmentation for video object tracking

Detection of moving objects in a video sequence is the fundamental step but a critical task in information extraction for computer vision applications. It provides focus on recognition, classification and analysis problems making the subsequent steps more efficient. Background subtraction, a common approach identifies moving object from video frame that differs from background. We propose an approach based on neuronal mapping for segmentation of targets with hybrid background subtraction and adaptive mean shift filtering. With this method, scenes containing moving backgrounds and the robust illumination changes can be considered effectively. First, the preliminary motion analysis is held to each block of the frame and the block with moving objects are detected. After thresholding and post processing the objects are obtained. Our method can handle scenes with moving objects and suitable for different types of videos. As our method supports inherent parallelism, it can be extended in real time.