An object-based algorithm for surveillance video compression
Date of Issue2008
School of Electrical and Electronic Engineering
The amount of video data generated by the security surveillance cameras is stupendous as cameras track peoples activity in a number of places like university campus, shopping malls, office place etc. There is a huge need to archive these videos, organize them effectively for security purpose and also for further analysis and research purpose. Because of the volume of data, video data needs to be compressed to reduce storage space required by them. This project is a step in the direction of compression of surveillance videos. The initial part of the project was aimed at obtaining a good segmentation of objects (people) in the security video. Perfect segmentation of moving people in the video was a challenge because of moving cast shadows and hence the project was directed towards eliminating shadows in objects during segmentation Different techniques were studied and an optimum object segmentation using adaptive threshold was proposed, implemented and tested. Better segmentation led to more accurate object-based motion vector estimation. The second part of the work involved the coding of the residual error object which is obtained by subtracting the original frame and the motion compensated frame. The theory of compressive sensing was studied and was used to code the error object, because of sparse representation of the error object. Different techniques of implementation of compressive sensing for error coding are discussed and compared Compressive sensing based coding was found comparable to the usual shape adaptive transform coding techniques. This report consolidates the steps involved in each stage in implementation of the compression of surveillance video and with comparative studies between different techniques.
DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing