Design And Realization Of The Parallel Tracking Control System By Client/Server Mode For Bio-mimetic Robotic Fish Based On Global Vision

Based on bio-mimetic robotic fish as the research object, this paper aimed at the researches on the real-time motion tracking method by multi-robotic fish. In the bio-mimetic robotic fish based on global vision, the robotic fish perceive the main source information from the position information and target information by tracking target motion, and the motion control of the robotic fish strategy depends on the perception of global visual information.This article uses the method of template matching tracking location of robotic fish, and focuses on the dynamic parallelization method of tracking algorithm to ensure that meet the real-time(<50ms) control robotic fish requirements.Based on it, design and implementation the MURobot tracking parallel control system of the C/S mode of global vision bio-mimetic robotic fish, and implements control strategy sent directly CPG parameter control robotic fish protocol extensions. Finally carried out the study on unsupervised depth self-encoding based on the deep learning of the biomimetic robotic fish, achieved the desired experimental results.The main research results can be summarized as follows:First, according to the requirement of the global vision system for image acquisition camera, vision subsystem selection of research using Mercury series of Daheng industrial cameras, image resolution and guarantee continuous long-term stability. The camera image by secondary development, continuous acquisition and parameter control, and converted to Open CV image formats.Second, using the method of template matching to identify location tracking robotic fish, combined with the robotic fish maximum speed and turning speed in local area within the scope of the rectangular search matching the robotic fish posture, and using the elliptical area detecting further improve matching accuracy and efficiency.Third, in multi-target tracking biomimetic robotic fish, in order to meet real-time(< 50 ms) control requirements, optimize the realization of template matching tracking algorithm parallelization. In the multi-target tracking, in order to fully exploit the advantage of multi-core CPU computing resources and with the current tracking number, using dynamic programming methods to achieve multi-target tracking algorithm for dynamic parallelization. Parallel algorithms using Open MP parallel programming, supports up to 16 tracking control of robotic fish at the same time. In the actual experiments, the tracking number of usually less than 8, the dynamic parallelization way makes the MURobot is compatible with better adaptability to reduce the hardware configuration requirements.Fourth, with the support of CUDA programming GPU hardware popularization and computational performance improvement, studies using CUDA open a large number of threads to realize tracking parallel computing speed up effect is obvious.Fifth, the design and implementation of the C/S model of global vision MURobot biomimetic robotic fish parallel tracking control system. System consists of Server, Client and write policy control of three parts. Robotic fish policy control according to the template by the user requirement and interesting to write and compilation, loaded from outside the system sending commands to perform control robotic fish movement. In the send command control, implemented in the control strategy directly send CPG parameters control of robotic fish protocol extensions.Sixth, carried out the study on unsupervised depth self-encoding based on the deep learning of the biomimetic robotic fish, achieved the desired experimental results. After training self-coding network has a strong ability of feature abstraction, continuous tracking estimate motion position of the robotic fish, and can adapt to different illumination change environment and the interference of other moving objects.Finally, the summary and analysis of relevant research work done herein, and pointed out that the global visual tracking bionic parallel machine control system in the direction of future research and application prospects.