DocumentCode :
3454513
Title :
Acceleration of pedestrian detection algorithm on novel C2RTL HW/SW Co-design platform
Author :
Zhu, Yihao ; Liu, Yongpan ; Zhang, Daming ; Li, Shuangchen ; Zhang, Pei ; Hadley, Tedd
Author_Institution :
EE Dept., Tsinghua Univ., Beijing, China
fYear :
2010
fDate :
21-23 June 2010
Firstpage :
615
Lastpage :
620
Abstract :
Vehicle-based pedestrian detection systems play a very important role in the field of intelligent transportation. However, existing pedestrian detection algorithms are too computing extensive for traditional vehicle-based processor. Considering the performance and flexibility of hardware/software co-design architecture (CPU+FPGA), it is feasible to implement these algorithms on such platforms. In this paper, we present a popular template-based detection algorithm on such a novel platform which including an ATOM processor and a FPGA-based accelerator, and designed by a novel C2RTL automatic design flow for the platform. Furthermore, we propose performance models for the design and explore configuration parameters of the platform for performance optimization. The experimental results show that our design can achieve nearly 20x speedup for the critical function block compared with the processor-only solution to meet real-time requirement.
Keywords :
field programmable gate arrays; hardware-software codesign; object detection; traffic engineering computing; ATOM processor; C2RTL automatic design flow; C2RTL hardware-software codesign platform; CPU; FPGA based accelerator; intelligent transportation; template based detection algorithm; vehicle based pedestrian detection system; vehicle based processor; Acceleration; Algorithm design and analysis; Computer architecture; Detection algorithms; Hardware; Intelligent transportation systems; Intelligent vehicles; Software algorithms; Software performance; Vehicle detection;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Green Circuits and Systems (ICGCS), 2010 International Conference on
Conference_Location :
Shanghai
Print_ISBN :
978-1-4244-6876-8
Electronic_ISBN :
978-1-4244-6877-5
Type :
conf
DOI :
10.1109/ICGCS.2010.5542990
Filename :
5542990
Link To Document :
بازگشت