Visual-Based Simultaneous Localization And Mapping Of Underwater Bionic Robot

The vast sea area not only contains abundant resources,but also is one of the main channels of trade and cultural exchange.The construction of offshore platforms and cross-sea bridges is one of the necessary means for developing the ocean.Regular inspection of the base pipes and piers of such buildings can reduce safety risks and create economic value.Bionic crab-like robot is an amphibious and multi-movement unmanned robot,which has the advantages of flexibility,stability and intelligence.It can complete the task of inspecting marine equipment,and can also be competent for scientific research and exploration in various underwater environment in the future.The objective of this thesis is to study the Simultaneous Localization and Mapping(SLAM)of bionic crabs in an unknown submarine environment.SLAM is different from the traditional underwater robot positioning,which is based on inertial navigation and assisted by other equipment.SLAM only relies on the environmental sensor,it can locate autonomously and construct the feature map of the environment by continuous observing the surrounding environment.In this thesis,the underwater camera is used as the sensor of environmental awareness,which has the advantages of low cost,small size.The camera can obtain color images,these images can be used to realize visual positioning and provide necessary information for detection and other tasks.The specific contents of this thesis are as follows:Firstly,the characteristics of underwater imaging are analyzed,and the methods of underwater image enhancement and recovery are introduced.In view of the feature loss and color information distortion of traditional enhancement algorithms,the Retinex theory is improved from two aspects:filtering and color space.Experimental results show that the improved enhancement algorithm can improve the quality of underwater images,and the enhancement process can also improve the efficiency and accuracy of positioning and mapping.Secondly,considering the features of simultaneous localization and mapping based on feature method and direct method,a hybrid SLAM method based on feature assistance is proposed.The hybrid SLAM method takes the direct method as the front end.Only when the direct tracking fails,the feature points will be matched,and the indirect feature tracking is used to recover the robot's pose.When these conditions are no longer violated,return to the direct method front end.In terms of mapping,some strategies for constructing a sparse feature point cloud graph are implemented,such as eliminating unreliable points according to the co-vision relationship.Finally,the hybrid SLAM method proposed in this thesis is tested on two open source land dataset,and the results show that this method can well adapt to various environments and unstable motion.We also used a ROV(Remotely Operated vehicles)to simulate bionic robot to conduct pool experiments,a feature point map of the pool was constructed and the reliability and effectiveness of the proposed algorithm has been proved.