Algorithms of video frame segmentation and video compression for video surveillance systems
Abstract
Algorithms of video frame segmentation and video compression for video surveillance systems
Incoming article date: 15.12.2014Article is devoted to solving the problem of video compression for panoramic video surveillance system. The actuality of the problem associated with both the limitations of existing solutions, as well as the peculiarities of video streams, which in many codecs are not counted. The novelty of the work is related to the fact that the role of the key frame is performed by image of background estimation (not by a frame from input video stream). In this paper we used the ideas of the four mainstream approaches in the field of compression (segmentation (division into a dynamic component and a static background), JPEG-compression, reduce the frequency of transmission of the reference frame (image of background estimation), the ability to change the resolution of the frame, the synthesis of the output image). The solution of problem of compression is produced by division of video frames into separate semantically meaningful elements (objects, background) transmitted over a communication channel with varying degrees of detail depending on the semantic significance and accompanied by metadata. It is experimentally proved that the value of the compression ratio of the video stream K = 135 times at peak signal to noise ratio PSNR = 32 dB, which allows to send the panorama of resolution 3072x720, stitched from the three cameras, through the channel with a capacity of 10-11 Mbit / s.
Keywords: video compression, segmentation, image processing, video surveillance, metadata