The Research And Implementation Of HIPs Image Matching Algorithm

With the development of society, objective tracking plays a more and more importantrole in both military and civilian. As one of the tracking method, Image tracking method isattracting people’s attention with its unique advantages. Therefore, research on image trackingtechnology and development of real-time tracking system are of great significance. In thispaper, we study a new image matching algorithm (Histogrammed Intensity Patches). Combinewith the DSP optimization technology, we realize the algorithm in the embedded platform andget a good real-time tracking result.In this paper, we study the existing image matching algorithms and find that they can beroughly divided into three categories，they are Grayscale-based image matching, descriptorfor invariant feature based image matching and classification-based image matching. Wecompare and analysis the characteristics of the three kinds of image matching algorithms, thenfocus on studying a new image matching algorithm based on descriptor for invariant feature.This algorithm uses an independent histogram of quantized intensity for each pixel location inthe patches, the author refer to this model for a feature as a Histogrammed Intensity Patch orHIP, and the algorithm called HIPs (Histogrammed Intensity Patches). Compare with thetraditional tracking algorithm, HIPs algorithm is relatively simple, and it can greatly reducethe amount of computation in matching process, which is quite suitable for the developmentof embedded tracking system. In the course of the research, we use MATLAB software andwrite code to realized the algorithm and verify the effect. We try to use different kind ofimages and different parameters in the training phase. According to the matching result ofdifferent training result, we summarize which factors can affect the matching results. We alsotry using different implementations in the interest point detection phase and homographyestimate phase to achieve the best matching result.Then the paper analysis the DM642DSP embedded experimental platform, including itsdevelopment environment, DSP chips structural features, the frame structure and every modelfunction of the system. According to the characteristic of the HIPs algorithm, we transplant itinto the DM642hard ware platform. Due to the programming differences between MATLABand DSP, we need to adjust the format of MATLAB training result. Variety of DSPoptimization technology is used to improve the efficiency of the tracking system, and finallywe obtain a good and real-time tracking result.