Optimization of processing time for vision based one-channel around view monitoring system on embedded platform

Abstract

Around View Monitoring (AVM) system is widely used as a part of driver assistance system. AVM helps driver to see blind-spot region and drive through narrow space by providing a 360-degree sight of the car’s surrounding from the top view. The project will implement a vision based one-channel AVM system on embedded board and focused on the algorithm optimization to improve the system’s processing speed. Two embedded boards, Telechip TCC 8971 SoC (Cortex-A7 quad-core) and Renesas R-Car H2 (Cortex-A15 quad-core) are used for this project and a PC platform of Intel i7 core is used to benchmark the processing speed of running the same algorithm on PC. Embedded board has a limited processing power and memory compared to standard PC platform. The common challenges of running an image processing on embedded board is that image processing requires high processing power and large memory to support pixel by pixel computation. The research objective focused on identifying the issues of running the AVM system on embedded board and implements several optimization methods to improve the initial processing time of the system. The thesis explains the use of Yocto Project to setup the embedded development board and algorithm porting. Several techniques of optimizations are presented in this thesis including multi-threading and the use of Qt OpenGL for GPU-accelerated output image rendering. The final optimization improved the total processing time of AVM system by 54.29 % on Telechip board, 52.04 % on Renesas board and 61.64% in PC platform.