People tracking in video

Abstract

It is very easy for humans to track a moving object or people in a video clip and to further analyze it to obtain meaningful information. However, it is difficult to get a computer to perform what humans can do. Furthermore, before computer can do analysis to human behaviours, it must track the human in the first place.

With the fast evolving technology nowadays, computer vision has taken to a greater height. More sophisticated surveillance systems are developed and more intelligent requirements are demanded. For example, identify potential terrorist in public places like railway stations or spotting enemies in military context to facilitate the planning of battle order. This lead to a key factor: people tracking.

The main focus of the project is to explore the latest state of the art techniques in visual tracking, and investigates its performances with respect to people in real home surveillance videos. Methods such as background subtraction, image morphology, and histograms will look into as it can be implemented to track people.

Further research are done to the methods mentioned in the proposed paper, “Robust Real-Time Visual Tracking using Pixel-Wise Posteriors” [1] such as region based segmentation, level set method, and Lucas-Kanade method.

Discussion on how the proposed paper implemented the idea will elaborated based on author understanding. Comparisons will then make between the proposed paper ideas and home surveillance videos.

The effectiveness of a surveillance system, entertainment devices like Kinect, behaviours analysis of humans is determined by its accuracy to track people.