Acceleration of pedestrian detection algorithms on a multi-core vehicle computing platform

• Published in: 2009 IEEE Youth Conference on Information, Computing and Telecommunication pp. 208 - 211
• Main Authors: Yongpan Liu, Yihao Zhu, Sunny Zhang, Senjie Zhang, Huazhong Yang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.09.2009
• Subjects: Acceleration; Computer architecture; Detection algorithms; Energy efficiency; Intelligent transportation systems; Intelligent vehicles; Multi-core; Multicore processing; OpenMP; Pedestrian Detection; Road safety; Road transportation; Vehicle Computing Platform; Vehicle detection
• ISBN: 1424450748; 9781424450749
• DOI: 10.1109/YCICT.2009.5382388

Vehicle-based pedestrian detection system receives more and more attentions in road safety applications of the modern intelligent transportation system. However, the existed detection algorithms are too computing extensive for single core vehicle-based processors. As the promising multi-core architecture provides both energy efficient and powerful computing solutions, it is relevant to evaluate the up-to-date pedestrian detection algorithm on such novel platforms. This paper implemented a popular template based pedestrian detection algorithm on such platforms and gave out computing bottlenecks by profiling. Furthermore, we proposed an application aware multithread accelerating technique. Experimental results showed that our design can achieve nearly 2 - 4x speedups on multi-core processors for critical function blocks, which reduces total execution time ranging from 26 to 41%.