Design And Implementation Of Real-time Motion Analysis System Based On ZYNQ Heterogeneous Structure

The video-image based target motion analysis is one of the most important research area of computer vision field, with vast application prospects in the various fields of national defense, military, scientific research, video surveillance, virtual reality, medical and other fields. Usually implemented computer vision video information processing system of the underlying hardware processing power has a very high demand. At the same time, it is a very difficult challenge to achieve system miniaturization, low power consumption, real-time processing capability and high robustness. The traditional general purpose CPU execution to achieve software-only based on building complex mathematical model and data analysis algorithm to meeting performance is difficult to realization, besides using ASIC implementations has a flaw that is the high cost and low expansibility. In order to solve the above problems, this paper makes a research and design in the following three aspects. First, I refer to the visual perception mechanism for the ventral stream of visual cortex, taking a typical visual perception hierarchical computing model as calculation model. Therefore, I design a spatial and temporal movement towards human action feature recognition algorithm that easily achieve on system-on-chip. Second, using a top-down system-level designs flow for recognition algorithms to optimize hardware and software division of tasks and implemented in a heterogeneous platform of ARM+FPGA. Third, using high-level synthesis method to designing a pipeline-type motions information extraction circuit dedicated hardware acceleration, so the hardware system can achieve 1080p@60fps HD video real-time image processing. Experimental data shows that based on the same algorithm in heterogeneous platform compared to a general-purpose CPU implemented, its processing speed increased 38 times; and then compared with the ASIC, which in turn reduces the difficulty of design and development, and enhance design scalability and flexibility.