Research On Image Processing System Based On Multi-core DSP TI-C6678

With rapid development of image acquisition devices, high-resolution image bring largeamount of image data and complex algorithms bring large amount of calculation. In order toachieve fast and intelligent image processing, image processing platform requires a higherprocessing performance and lower power consumption, so the DSP of high-speed andhigh-performance is critical for realizing rapid processing. Based on the purpose of improvingperformance of image processing platform, this paper focuses on the research of multi-core DSPTI-C6678to solve the balance problem between perfomance and power consumption anddesigns parallel model to play the multi-core performance advantanges, which accelerate theschedule of high-performance image processing system.This paper regards exerting full multi-core performance as the starting point, and carriesout respectively in-depth research from the following aspects. First, we propose an optimizedapplication scheme of memory to enhance core access efficiently by analysing its principle andarchitecture. Meanwhile, research on the resource sharing and interrupt management mechanismupgrades system’s coordination and flexibility. Second, inter-core synchronization andcommunication plays a pivotal role for image processing system. We summarize four techniquesand advance the optimal scenario based on their own merits. Third, in view of the deminds forbig data rapid interaction between memory on our platform, we design an optimized scheme forhigh-speed data interaction based on DMA technologies，realizing GBps level performancewithout CPU intervention and having both high flexibility and advantage of communicationsbandwidth. Fourth, the image processing platform must have the engineering requirements ofboot startup, researching and realizing multi-core boot startup based on SPI. At the same time,taking into account the network expansion, this paper completes network communication basedon TCP/IP protocol which achieving real-time network performance and enriching the datainterfaces. Finally, considering the difficulty of multi-core development environment based onTI-C6678, this paper investigates multicore development framework of master-slave and data-flow mode, analyses its mechanism and methods of synchronization and communication to makea right decision for selecting the most appropriate framework for image processing algorithms.Combination of the above analysis, this paper constructs the image processing models ofmulti-core DSP serial and multi-core DSP parallel based on our former conclusions of memoryoptimization, high-speed data interaction and inter-core synchronization mechanism. Themulti-core serial model uses sequential processing structure similar to single core, has a greaterloss of time and can not bring full multi-core performance into play. The multi-core parallel model improves the triggering mechanism between cores of multi-core serial, advances aninnovative back-drive means to trigger other cores. This model makes cores execute imageprocessing at the same time and sloves the problem of data coverage, which polish up the coreefficiency, compress powerfully entire task’s processing time, strong real-time ability and attainthe aim of upgrading the processing performance for multi-core image processing system. Inaddition, multi-core parallel model has better scalability and greater performance improvementspace.