Image processing and DSP technology applied to remote activity monitoring : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Technology at Massey University

Abstract

This thesis describes the development of image processing algorithms to achieve industrial remote activity monitoring. The design of such algorithms is constrained by the requirement that they must be implemented in real-time on a cost effective digital signal processor (DSP). A review of the literature revealed that two viable alternatives were available. These are JPEG, a well established lossy block-based image compression algorithm and methods using wavelets. A suitable family of wavelets was identified as the biorthogonal-7.9, the coefficients indicating the number of taps in the quadrature mirror filters at the heart of the transform. Data compression is then achieved through the construction of zerotrees followed by sequential baseline coding. The two candidate algorithms were then systematically compared for recovered image quality and implementation cost at high compression ratios in the range of 16:1 to 64:1. On this basis the wavelet approach was selected and its implementation on a DSP studied. The architectural features of the Motorola 56303 DSP are presented and analysed. It is shown that the various components required for the wavelet based algorithm can be efficiently mapped onto the DSP architecture. Motion detection and image watermarking algorithms were designed and co-operatively implemented with the compression algorithm. A new method of watermarking highly compressed images was developed and this algorithm has been named the Image Authentication Watermark. A new way of representing optimal Huffman code tables has been developed to enable Huffman entropy coding to perform competitively with the more complex arithmetic coding. A product of this research is a smart digital camera that has been integrated into an automated video surveillance system now in industrial production.