Research And Implementation Of Multi-Core Processor Parallel Processing On Omnidirectional Vision

Recent years, with the rapid popularity of multi-core processors, multi-core parallel computing has became a hotspot in computer science. Although the compute power of multi-core processors enhanced significantly, most software currently was developed for single-core processor, which couldn't utilize the advantages of multi-core processors. In order to make use of the multi-core resources effectively, this article proposed an approach to make the traditional serial process software running parallel on the multi-core processors, and it was applied into all-round visual detection and tracking system.Firstly, this paper made an analysis and comparison among multi-core parallel programming, single-core processor programming and multi-machine distributed programming. And design a procedure for making the serial process software running parallel on the multi-core processors. Secondly, because the cost of communication among nuclear on multi-core processor was less than that among several computers, so we improved the DAG model based on this. After analyzed the dependencies between the tasks, the dependencies between the data and the dependencies between procedures, this paper proposed the DAG graphics structure. In the light of the advantages of less cost of communication and shared memory, it improved list scheduling algorithm and achieved the task scheduling of DAG graph.Based on the analysis of detection and tracking task in system, this article divided the task into several sub-tasks with interdependent relationships. And proposed a parallel algorithm based on DAG model, this paper also constructed detection and tracking system based on this, the improved task scheduling algorithm was applied to realize task scheduling of DAG graph in the detection and tracking system.Finally, after achieving parallel design on the full of visual detection and tracking system, this article implement the enhanced visual detection and tracking system on dual core process. The whole detection and tracking system can have a acceleration of 1.55, and the best part can have the ratio of speed up to 1.71; The task load detected by Intel Thread Profiler was well balanced on the two processors.