Object detection and tracking in video image

Abstract

In recent days, capturing images with high quality and good size is so easy because of rapid improvement in quality of capturing device with less costly but superior technology. Videos are a collection of sequential images with a constant time interval. So video can provide more information about our object when scenarios are changing with respect to time. Therefore, manually handling videos are quite impossible. So we need an automated devise to process these videos. In this thesis one such attempt has been made to track objects in videos. Many algorithms and technology have been developed to automate monitoring the object in a video file. Object detection and tracking is a one of the challenging task in computer vision. Mainly there are three basic steps in video analysis: Detection of objects of interest from moving objects, Tracking of that interested objects in consecutive frames, and Analysis of object tracks to understand their behavior. Simple object detection compares a static background frame at the pixel level with the current frame of video. The existing method in this domain first tries to detect the interest object in video frames. One of the main difficulties in object tracking among many others is to choose suitable features and models for recognizing and tracking the interested object from a video. Some common choice to choose suitable feature to categories, visual objects are intensity, shape, color and feature points. In this thesis, we studied about mean shift tracking based on the color pdf, optical flow tracking based on the intensity and motion; SIFT tracking based on scale invariant local feature points. Preliminary results from experiments have shown that the adopted method is able to track targets with translation, rotation, partial occlusion and deformation.