Research On Autonomous Visual Localization Algorithm Of Amphibious Robot In Field Environment

Amphibious robots can directly serve many applications,e.g.the aquatic products,environmental exploration,and disaster rescue,etc.It is an effective way to improve China's amphibious related industries and national defense capabilities.However,the efficient and autonomous operation of amphibious robots in field environments relies heavily on the stable and accurate localization ability.In GPS-denied environments,traditional solutions to achieve autonomous localization is to utilize visual odometer technology through a sequence of consecutive images acquired by the camera.However,for amphibious robots operating in field environments,the motion characteristics,undulating terrain and complex lighting conditions cause many problems including lack of visual information and lost of field of view.The problems prevent the traditional localization algorithms from directly applying for amphibious robots.To achieve an accurate and stable visual localization system for amphibious robots in the field environments,the thesis designs an amphibious active visual inertial odometer system(Amphi Active VIO).The system is designed based on a tightly coupled visual inertial odometer with a sliding window.The image acquired by the camera and the data measured by the inertial measurement unit are fused to estimate the pose of the amphibious robot in real time.At the same time,to solve the problem of lost of field of view,an optimal viewing angle is calculated by using the perceived scene feature points and the current motion direction of the robot.The optimal viewing angle ensures stable visual localization while maximizing the exploration of the scene.Then,an active exposure control algorithm is developed to solve the missing problem of visual information caused by complex illumination conditions in field environments.The algorithm calculates the most suitable exposure time at the next frame by maximizing an predefined image gradient metric to ensure that the next frame is not overexposure or underexposure.Finally,to solve the problem of feature tracking lost caused by the robot's motion,several additional visual constraints are obtained by matching the newly detected feature points with the local map,and the method effectively increases the stability of the visual inertial odometer.To verify the effectiveness of the proposed system,the localization algorithm basedon the optimal viewing angle was first verified in a simulated environment.The algorithm ensures the stable localization while achieving the maximum exploration of the scene.Then by equipping a three-degree-of-freedom gimbal on the amphibious robot,it is verified that the active view adjustment prevents the lost of field of view during climbing motions,and performs much better than that with the fixed view solution.Further,experiments on the active exposure control(AEC)algorithm in the fields under high dynamic lighting environments show that the active exposure control can preserve more image details than the automatic exposure of the camera.The AEC algorithm achieves stable feature tracking,and improves the accuracy of the visual localization algorithm.Experiments on the public data set Euroc also show that the developed Amphi VIO with additional feature matching can improve the accuracy of visual localization compared to Vins-Mono.Finally,the effectiveness of the whole Amphi Active VIO localizaiton system proposed in the thesis is verified with our amphibious robot platform in actual field environments.