A Parallelism Analysis Pattern for Real-time Vehicle Recognition Algorithm

With the rise of many-core, parallelization is expected to be the mainstream to speedup the computation. This paper presents the study of classic vision based algorithms applied to vehicle recognition in DAS domain from the point of view of building a practicable real-time system. The workloads of the algorithm is illustrated and analyzed. After summarized the difficulties prevent current algorithm from achieving real-time, especially considering recent many-core tide, this paper brings out a parallel analysis pattern for vehicle recognition, our prime work indicates it can effectively help us to most exploit and express the parallelism of the algorithm.