Research On Multi-source Environment Modeling And Localization Technology Of Mobile Robot Based On V-SLAM And Panoramic Map

The mobile robot is equipped with multiple sensors to collect multi-source information for map construction and positioning,which lays a foundation for the realization of unfamiliar scene perception and understanding.Aiming at the Positioning problem of complex environment in the absence of GPS(Global Positioning System),this paper carried out research on the modeling and positioning technology of mobile robot multi-source environment based on V-SLAM and panoramic map.An innovative method combining multi-source image fusion and 3D rendering is proposed to realize multi-source panoramic map construction.Using the perceptual hash image retrieval algorithm,the RGB image matching method is established in this paper to achieve the multi-source panoramic map coarse positioning and visual SLAM precise positioning technology.Firstly,Kinect V2 camera was used to create an RGB-D three-dimensional dense point cloud V-SLAM map using ORB feature extraction method.The internal and external parameters of Kinect V2 camera were calibrated,and the internal and external parameters of depth camera and color camera were obtained.By combining ORB with RANSAC algorithm,feature extraction was carried out to realize image matching between multiple frames.At the same time,Pn P and BA were optimized to estimate the local camera pose.Finally,loopback detection was used to optimize the pose and construct a three-dimensional dense point cloud V-SLAM map efficiently.Secondly,aiming at the problems that the traditional single-source panorama provides limited information,is susceptible to illumination interference,difficult to capture hidden objects,and poor 3D experience for users,the multi-source 3D panoramic map is constructed by using multi-source image fusion and three-dimensional rendering environment modeling methods.The mobile robot is equipped with motion control module and multi-source camera,and carries out image acquisition in six directions,namely front,back,left,right,upper and lower,to obtain infrared images and color images.Hue,Intensity,Saturation and wavelet transform techniques are used to carry out image fusion.The cube texture mapping was carried out with Open GL rendering pipeline,and then the sky box model was used to simulate the real natural scene to improve the visual effect,and the scene map with realistic sense was generated to realize the construction of 3D panoramic map.In addition,a variety of image sharpness evaluation methods were adopted to carry out the resolution evaluation experiment on the multi-source panoramic map,which verified the effectiveness of image sharpness improvement.Finally,a multi-source panoramic map matching method and RGB image matching method are established by using the perceptual hash image retrieval algorithm,and the multi-source panoramic map coarse positioning technology and visual SLAM precise positioning technology are realized respectively.The perception hash algorithm is used to convert the current frame image and the panoramic map library constructed at fixed points into spectral map by DCT,calculate the hash value,generate the image fingerprint,and calculate the image similarity,thus completing the rough positioning process.After arriving near the fixed point by using rough positioning,precise positioning can be accomplished by using ORB-SLAM2three-dimensional dense point cloud map with the help of pose information in V-SLAM map,it avoids the accumulation error of traditional large-scale point cloud map construction to achieve precise positioning.In the experimental part,the sensitivity of perceptual hashing is evaluated by using several similar images,and the effect of perceptual hashing is evaluated by using a variety of similarity evaluation methods.The experiment also shows the practical feasibility of realizing the coarse location algorithm of panoramic map by using perceptual hashing.At the same time,the three-dimensional dense point cloud map near the fixed point is constructed to provide conditions for the accurate navigation in the future.