A Study on Data Parallel Optimization for Real-time Vehicle Recognition Algorithm

Data parallelization is an important method for speed up the vision based vehicle recognition system. And with the risen of multi/many cores, new programmability has been introduced to extend its ability to express more complicated and irregular algorithm utilizing multi/many cores hardware, this paper presents our study on data parallel computation model and illustrates our prime data parallel optimization for the key algorithms of vision based vehicle recognition algorithm including image filters, classifier and motion estimation. And after analyzing the result of optimization, the applicably bound of data parallel optimization and the future direction of parallel optimization for vision based vehicle recognition algorithm is summarized.