Research On Robust Visual SLAM Based On ORB In Complex Scenes

As the application scenarios of mobile robots becomes more and more widespread,all kinds of robots have gradually entered people's field of vision.The technology of SLAM makes a contribution for mobile robots to explore unknown spaces and perform precise navigation by relying on their own sensors.The existing SLAM methods are mainly designed based on ideal scenarios.There are many factors that affect the stability of the SLAM system in a complex environment.This article mainly focuses on complex non-ideal environments such as lack of environmental textures and unstable lighting scenes in real complex environments.The visual odometry and loop closure detection modules in visual SLAM are researched and improved.1.Aiming at the problem that the lack of object texture in the environment lead to the loss of visual odometry tracking,a strategic fused method of the direct method and the feature point method of the visual odometry is proposed.Based on the complex scenes with different focus aspects,the best method to calculate the camera pose is reasonably selected.Aiming at the situation that the feature points are difficult to extract due to the lack of texture,the direct method is selected to calculate the pose.According to the number of the feature points that are effectively matched,the preliminary results of another method are selectively integrated into the direct method,and the camera pose is optimized by joint error to obtain more accurate results.For other ordinary scenes with richer textures,the feature-point-method is selected to solve the camera pose.Experiments are designed to compare the performance of the algorithm in this paper with other algorithms.Verification by comparative experiments,the robustness of the fused visual odometer in complex environments has been improved.2.Aiming at the problem of under-exposure or over-exposure of image caused by changes in lighting in the environment,thereby reducing the recall rate of loop,the traditional loop closure detection method based on the visual bag-of-words model is improved.First of all,a module that can delete and add visual words online is added to the bag of words model,which can realize the establishment of a visual bag of words online,and keep the number and quality of visual words in the bag of words stable.Secondly,a preprocessing method has proposed to make the algorithm robust in the environment with obvious outdoor changing light.Through the detection of image exposure,the overexposed and underexposed images are preprocessed.Next,the images are passed to the module of loop closure detection for processing.The algorithm in this paper is tested with other mainstream algorithms.The experimental results show that the improved algorithm is more robust in outdoor scenes with variable lighting,and can meet the real-time requirements of loop closure detection.Finally,the improved loop detection algorithm,fused visual odometry and back-end optimization are combined to obtain the FD-fusion-lcd system.After experimental verification,the performance of the system is better than pure visual odometry positioning methods and pure schemes of SLAM based on feature point method.The effectiveness of the algorithm in this paper is verified.