Multiple Target Detection Using Optical Flow

Abstract

In this paper, a new sequence will technique for calculating the optical flow field between successive images of a moving scene which contains a single target or multiple targets is presented. Optical flow can be defined as the displacement estimate vector between corresponding regions of two successive images in a moving image sequence. To obtain the optical flow field, successive images in an image sequence are divided into small blocks of equal sizes. A row of pixels of the previous image block is adaptively tracked by means of the gradient of a similarity function, thus obtaining the displacement estimate components for that block. Then the displacement estimate components are combined vectorially to obtain the optical flow field. The stationary background always appears to have constant motion with respect to the sensor, which produces a constant optical flow field. Target motion produces an anomalous optical flow pattern. Detecting target motion is equivalent to detecting the anomalous optical flow. The optical flow fields are calculated for real image sequences and it is demonstrated by using these actual images that the algorithm performs with considerable accuracy.