Optimization And Application Research Of Visual Odometry Based On Image Enhancement

With the rapid development of the field of artificial intelligence,people's requirements for the automation of products are getting higher and higher.The development of mobile robots has brought great convenience to people's work and life,among which being able to complete autonomous navigation and localization is a prerequisite for performing tasks in complex environments,and at this time the importance of simultaneous location and mapping technology is reflected,while visual odometry as the front end of SLAM system,which is the basis for mobile robots to complete autonomous navigation and localization,so research on visual odometry has a great importance.However,in the actual operating environment of the robot,there are often too bright or too dark illumination,and it is due to the uneven brightness of the ambient illumination that the acquired image details are lost,and these low-quality images affect the subsequent extraction and matching of feature points,which eventually leads to inaccurate calculation of the robot trajectory.To address this problem,this dissertation adopts Gamma adaptive adjustment and detail enhancement algorithm to enhance the quality of degraded images and improve the matching effect of feature points,the experiments are conducted to verify the feasibility of this dissertation's algorithm through data sets in four different environments,and verify the feasibility of the algorithm of this dissertation in a real environment by building a robot platform.Firstly,the images obtained by the camera are screened,and the images are classified into three categories: over-bright,over-dark and suitable brightness by setting thresholds,and the overbright and over-dark images are adjusted by Gamma correction function,but considering that different images need different degrees of correction,adaptive Gamma correction is proposed to realize adaptive adjustment of image brightness;Secondly,to further improve the detail information of degraded images,the multi-scale detail enhancement process is performed for the brightness-adjusted images to enhance the image detail information;Finally,the initial estimated poses of the mobile robot are obtained by the Pn P method,and the optimized estimated poses are obtained by nonlinear optimization.The experiments use TUM data set to test the algorithm of this dissertation,and evaluate the experimental results in terms of absolute trajectory error and relative rotation error.The experimental results show that after the processing of the algorithm in this dissertation,the ORB feature point extraction and matching effect of the image are effectively improved,the absolute trajectory error,relative translation error and relative rotation error are reduced,the trajectory accuracy of the mobile robot is improved to a certain extent.Meanwhile,this dissertation building robotic platform in real-world environment,the robot trajectory is consistent and smooth,which verified the feasibility and effectiveness of the algorithm in this dissertation.