Unknown Moving Target Detecting and Tracking Based on Computer Vision

In this paper, we applied inter-frame difference and optical flow algorithm to detect unknown moving target, and used the particle filter algorithm to track the detected moving target. Because the compute time of optical flow is long, we use inter-frame difference method to extract approximate target region and then calculate local optical flow. A dynamic elliptical template with affine transformations was constructed and a dynamic motion model was established to predict particle state. The complex moment was used as the feature in the reference region and candidate regions. At the same time the Gaussian function was used to calculate particle weights so that the particles with small weights were resampled. Finally, the tracking object state was computed by using particles weighted sum. Experiment results show that the speed of recognizing and tracking moving target is improved and the template can dynamically do some affine transformations with the moving target.