Study On The Real-time Embedded Visual Tracking System For The Bio-inspired Amphibious Spherical Robot

As the combination of driving mechanisms and automation techniques,mobile robots have great potential on applications such as search and rescue missions.Amphibious robots have become a kind of novel industrial equipments due to its advantages in environmental adaptability and mobility.The visual tracking system is a key component of amphibious robots to perceive the environment and to make behavioral decisions,which are critical important to realize intelligent functions such as motion tracking,visual servoing,and multirobot cooperation.Different from the conventional robotic vision applications in the terrestrial environment,the visual tracking systems designed for amphibious robots have to deal with interference factors such as the low imaging quality,the complex light conditions,the ever-changing observation view,and the fast motion of the target object.Thus,it is a very challenging work to ensure the precision and the robustness of the robotic tracking system.Moreover,because of the limited drive capability of available amphibious driving mechanisms and the application requirements of reconnaissance mission,most amphibious robots are small-sized and relatively weak in the capabilities of payload carrying,information processing,amphibious communication,etc.Therefore,the efficiency and the real-time performance of the robotic tracking system should be especially considered.Given the constraint conditions of robotic vision applications in the complex amphibious environment,this thesis tried to explore the feasible solution to designing an embedded visual tracking system for small-sized amphibious robots.The robot platform,the processing system architecture,and the vision algorithms were studied to fit in with the characteristics of amphibious robots and environments.First,an improved bio-inspireda mphibious spherical robot equipped with the “wheelleg-vectored thruster” compound drive mechanism and amphibious sensing devices were designed and constructed.The mechanical structure and the dynamical model of the robot were analyzed.Then an adaptive control system was designed and implemented considering the environmental disturbances and the system characteristics,which enables the robot to realize the land motions in varied terrains and the underwater motions in four degrees of freedom.Second,an asymmetrical and heterogeneous computing system architecture was proposed to construct the embedded vision system of the amphibious spherical robot which should be low-power,high efficient,multi-tasking,and portable.Under this architecture,customized hardware accelerators were designed and deployed to accomplish the arithmeticintensive processes of vision algorithms.Adaptive vision algorithms including MSRCR(Multi-scale Retinex with Color Restoration),Gaussian model-based moving target detection,and compressive tracking were implemented in real time using the proposed architecture for automatic robotic applications.Third,visual tracking algorithms were designed towards characteristics of the amphibious environments.On the basis of compressive tracking algorithm,theoretical tools,including the particle filter,the online updating appearance model,and the adaptive learning mechanism,were adopted to improve the precision and robustness of the tracking algorithm.Furthermore,a RGB-D imaging system was construct to acquire 3D information of the surroundings.By fusing the tracking results or appearance features of visual images and depth images,a robotic tracking algorithm,which was able to handle occlusion and scale changing problems,was designed and implemented.Experimental results demonstrated the adaptability and mobility of the improved amphibious spherical robot.Moreover,the proposed tracking algorithms and robotic vision system were capable of meeting application requirements of the amphibious spherical robot in terms of real-time performance,precision,and efficiency.By combining the adaptive motion control system and the visual tracking system,the robot was able to realize robotic functions such as the following motion,the motion tracking,and the robot formation.The study in this thesis may be important to improve the automation and intelligence level of small-sized amphibious robot.It may also be meaningful to promote the practicality of vision systems for mobile robots and underwater robots.