Research On Visual Location Technology To Adapt To Changes In Indoor Scenes

Visual location technology has been widely used in the field of artificial intelligence,and has become almost one of the indispensable functions of mobile robots.The study of the visual location of mobile robots in indoor scenes is one of the important research directions.Among them,the common lighting changes in indoor environment and the decrease of texture richness are easy to affect the visual location of mobile robots,so this paper studies and analyzes the lighting changes and weak texture environment in indoor scenes,and based on the monocity camera,a pure visual location scheme for indoor scene changes is proposed,which improves the robustness of visual location in lighting changes and weak texture environment while maintaining certain accuracy.In addition,a map-based tight fusion location algorithm is proposed,which effectively limits the growth of cumulative error.For the above content of this paper's main research work is as follows:1.Visual location algorithm based on direct and indirect methods.Based on the advantages and disadvantages of direct and indirect methods,this paper analyzes and discusses its robustness and location accuracy in weak texture environment and lighting change environment.On this basis,the advantages of the two methods are combined to build a mixed visual odometer.In this paper,the pixel features with obvious gradients and the corner features with high differentiation are extracted,and the polar search matching and depth convergence optimization between the images are constructed for the two features respectively,so as to ensure the efficiency of the algorithm while restoring the feature points to the depth information as far as possible.Finally,the photometric error is established for the pixel feature construction photometric match,and the veragonal point construction descriptor match is established to establish the re-projection error to optimize the position and map point depth nonlinearly.By measuring the Euroc data set and the IROS data set,the validity of the algorithm in the environment of weak texture and slight light change is verified.2.Online image photon correction algorithm.In view of the scale change and even failure of the visual location algorithm in the scene of the dramatic change of luscity,the response function model and vertigo model of the camera are constructed by the camera imaging process.The spatial position matching and photometric variation between images are obtained by sampling the LK optical flow tracking on continuous frames by the image feature points,and the energy functions of response function,halo factor,exposure time and scene irradiation are established,and the solution is solved by nonlinear optimization.The algorithm corrects the image luscity online to local consistent,effectively weakening the influence of the sharp change of lunscity on the visual location algorithm in this paper.3.Map-based visual location algorithm.The accumulation of errors in the long-term,long-distance operation of the SLAM system.Therefore,based on the practical application,this paper puts forward an algorithm based on visual map location.The algorithm uses the DBOW map to establish the same index as the feature descriptor of the image,and effectively uses this method to retrieve and match the map efficiently and establish the map-based relocation.In addition,this paper proposes a prediction method for local location maps based on the location position in slam process,and re-projects the successfully matched map points into the sliding window optimization of slam system,and establishes a map-based tight fusion location algorithm.The validity of the algorithm is verified by the actual test.