A CUDA based parallel processing for H.264 inter prediction decoder of car Management System

Autor: Bo-Yu Hou, 侯柏宇
Druh dokumentu: 學位論文 ; thesis
Popis: 101
Most of today's vehicle management system is still dominated by the traditional CPU serial decoding, however, along with the multiplication of transportation scale, the demand for vehicle management system will simultaneously to expand, so how to expand this system equipment will become a huge of problems. This paper presents a new type of vehicle management system includes the basic vehicle management system functions such as Channel track on map, the corresponding map image, and transplanted H.264 inter-frame motion compensation prediction part to the GPU to do parallel processing. Since there is no data dependency between each Channel problems, so do quite suitable for GPU parallel processing each channel. The demand for vehicle management system respectively will be introduced with the implementation of H.264 codec, Google Map API and CUDA parallel processing. In this study, due to the memory transmission bottlenecks between host side and device-side. And the inadequate planning number of scale parallel computing Block, The performance of experimental motion compensation will increase 12%-29% while handing channel number rise to 8-16 lens. However the overall performance of the vehicle management system is no significant improvement of using the CPU decoding.
Databáze: Networked Digital Library of Theses & Dissertations