An application specific low bit-rate video compression system geared towards vehicle tracking.

Abstract

The ability to communicate over a low bit-rate transmission channel has become the order of the day. In the past, transmitted data over a low bit-rate transmission channel, such as a wireless link, has typically been reserved for speech and data. However, there is currently a great deal of interest being shown in the ability to transmit streaming video over such a link. These transmission channels are generally bandwidth limited hence bit-rates need to be low. Video on the other hand requires large amounts of bandwidth for real-time streaming applications. Existing Video Compression standards such as MPEG-l/2 have succeeded in reducing the bandwidth required for transmission by exploiting redundant video information in both the spatial and temporal domains. However such compression systems are geared towards general applications hence they tend not to be suitable for low bit-rate applications. The objective of this work is to implement such a system. Following an investigation in the field of video compression, existing techniques have been adapted and integrated into an application specific low bit-rate video compression system. The implemented system is application specific as it has been designed to track vehicles of reasonable size within an otherwise static scene. Low bit-rate video is achieved by separating a video scene into two areas of interest, namely the background scene and objects that move with reference to this background. Once the background has been compressed and transmitted to the decoder, the only data that is subsequently transmitted is that that has resulted from the segmentation and tracking of vehicles within the scene. This data is normally small in comparison with that of the background scene and therefore by only updating the background periodically, the resulting average output bit-rate is low. The implemented system is divided into two parts, namely a still image encoder and decoder based on a Variable Block-Size Discrete Cosine Transform, and a context-specific encoder and decoder that tracks vehicles in motion within a video scene. The encoder system has been implemented on the Philips TriMedia TM-1300 digital signal processor (DSP). The encoder is able to capture streaming video, compress individual video frames as well as track objects in motion within a video scene. The decoder on the other hand has been implemented on the host PC in which the TriMedia DSP is plugged. A graphic user interface allows a system operator to control the compression system by configuring various compression variables. For demonstration purposes, the host PC displays the decoded video stream as well as calculated rate metrics such as peak signal to noise ratio and resultant bit-rate. The implementation of the compression system is described whilst incorporating application examples and results. Conclusions are drawn and suggestions for further improvement are offered.