An Approach to Segmenting Initial Object Movement in Visual Sensor Networks

This paper suggests a new method to extract the initial movement of moving objects in digital image data obtained in visual sensor networks. First, consecutive images are received as input. Then, the frames are partitioned into nonoverlapping square blocks of pixels, and finally, the block-based motion vectors, which represent the movement information between two adjacent frames, are extracted from the received images using a block-matching algorithm. The extracted motion vectors are subsequently applied to an outlier-elimination algorithm called robust estimation to discriminate between the background motion vectors and those of noise or moving objects. The motion vectors corresponding to the noise or objects are clustered with an unsupervised clustering algorithm to segment the individual moving objects. Experimental results prove that the proposed method can effectively detect the initial movement of objects in various indoor and outdoor environments.