Segmentation of moving objects in visible and thermal videos

Segmentation of moving objects is one of the fundamental tasks in computer vision applications. Many of the proposed segmentation algorithms work well for visible spectrum but fail to work for night vision thermal videos primarily due to the halo effect surrounding the object of interest. For video analytics and security applications, it is essential that the solution works round the clock or 24x7. In this paper we present an efficient technique for segmenting moving objects from visible and thermal videos. We present an approach that makes use of block matching algorithm for differentiating between background and moving objects. The full search or the exhaustive search being computationally expensive, we propose a threshold based strategy that reduces the computational complexity to a large extent to make the solution suitable for real-time applications. The reduced computational complexity makes it suitable for deploying on any low cost desktop computer.