The Research And Application Of Pedestrian Semantic Image Based On DM642

In the fields of the artificial intelligence and the pattern identification, the video image analysis is one of the most challenging and important work, the target of which is the achievement of the function of electronic facilities to imitate the system of human vision, so that electronic facilities have the ability of perceiving and understanding images like the system of human vision. Since the introduction of semantic images, we have made a great progress on this target, because the understanding of images hasn't been only confined to the level of the bottom. Although the high-level semantics of images is a new born child in the field of image manipulation, it will become one of the most important and popular research areas in this field.Formerly most of digital image process systems are based on personal computer, the PC is high consuming power and lower efficiency of the system; so it's not suitable for stand-alone systems. In recent years, the constant advancement in technology increased the processing speed of Digital Signal Processor (DSP) and expanded its functionality, DSP is highly accurate and efficient in processing high volume and complex algorithms, not only are DSP-based systems small size and suitable for stand-alone but also meet most systems' requirement for real-time performance. The main research area of the thesis including:(1) The construction of the terrace of Embedded Software Development. The design of the system according to the RF5 developed by the company of TI and XDAIS, using the model of Thread, Channel, Cell and XDAIS algorithm. Relevant clipping and allocation of DSP/BIOS embedded real-time kernel developed by the company of TI, adding performance monitoring modules.(2) The detection of the potential ROI (Region Of Interest) of semantics. We can obtain ROI through firstly transplanting two kinds of frequently-used detection algorithm of moving aims in the PC (the finite-difference time-domain method and optical flow) to DSP and then median filtering and Blob analyzing the result of the detection. After the experiments, the thesis estimates respectively the performance index of the algorithm of DSP of these two kinds, using maximum warehouse of load and system, and depth and maximum interrupt response time.(3) The matching of semantics of pedestrians. The thesis makes semantic formwork of pedestrians applied to this system, designs efficient algorithm of the matching of the inner product applied to that of semantics of pedestrians, making use of characteristics of directly concurrent computation of several multiplications of DM642 DSP and then analyses experimental result.The contributions made in this thesis introduce the concept of semantic image to embedded system and provide a step of progress on the way towards applying.