Theory Of Binocular Stereo Vision And The Applied Research In Gesture Recognition

With the development of information technology,3D scene reconstruction technology has obtained great progress over the last few years,and is widely used in the fields of virtual reality, robot obstacle avoidance, unmanned aerial vehicle flight and so on. As a natural way of human-computer interaction, gesture has a good user experience and gets more and more attention. The combination with stereo vision makes gesture recognition to be able to identify 3D gesture and the has a good recognition effect.Binocular stereo vision has advantages such as low cost, data acquisition easily and non-contact in the aspect of obtaining 3D depth information. Platforms such as Central Processing Unit (CPU), Graphics Processing Unit(GPU) and Digital Signal Processor(DSP) cannot meet the requirements of high-speed real-time image processing. Support for the high performance embedded system exist deficiencies. Field-Programmable Gate Array(FPGA) provides a large number of logical operation unit,and FPGA has unique advantages in parallel image processing. Zynq-7000 all programmable SoC integrates ARM Cortex-A9 processor and FPGA in one chip, achieve the perfect combination of serial processing and parallel processing. Efficient embedded solution provides a more flexible choice between performance and cost.In this paper, study and analyze the storage mode and operation mechanism of SAD firstly.By using the advanced algorithm modeling tool high-level systhesis(HLS),a high performance algorithm of SAD which applicable to the FPGA has been implemented.On this basic,a distinguish algorithm is improved aimed at the shortcoming of skin tone recognition which is difficult to distinguish face and hands. Finally, two algorithms are verified on the Zynq- 7000 SoC device, a embedded gesture recognition system based on stereo vision has been implementated.The experiment results show that the system has a good distribution of software and hardware resources,good man-machine interaction,high-speed processing performance and excellent network remote control function. And compared to other system,this system achieve the low cost and low power consumption, provides reference for further high performance image processing system.