Application Research In Mobile Robot VSLAM With Deep Learning

With the rapid development of artificial intelligence technology,intelligent robots have been widely used in industrial production,military operations and all aspects of people's lives.Whether it hns the obility of simultaneous localization and mapping becomes one of the key conditions to determine whether a mobile robot has the ability of autonomous navigation.It has been widely used in visoal-bassd simultaneons localization and mapping for mobile robots because the visual sensor has the advantages of low cost,wide detection range,large amount of information,rich features,and easy to extract image features.To solve the problem of low positioning accuracy of the robot and low accuracy of loop closure detection in VSLAM,the AKAZE algorithm is firstly applied in the feature extration and matching for the mobile robot VSLAM,and the 3D VSLAM algorithm based on AKAZE for the mobile robot is proposed to improve the positioning accuracy of the mobile robot.Then,the deep learning model Places365-VGG is applied in loop closure detection of the VSLAM.A loop closure detection algorithm based on Places365-VGG for the mobile robot VSLAM is proposed.The effectiveness of loop closure detection algorithm based on Places365-VGG for the mobile robut VSLAM is verified.The main achievements of this paper are as follows:(1)a high-precision three-dimensional VS'LAM algorithm for the mobile robot in proposed,that is the 3D VSLAM algorithm based on AKAZE for the mobiJe robot.The algorithm mainly includes two parts;the front end and the back end.The feature extraction and matching algorithm AKAZE is applied to the RGB image collected by the Kinect camera to establish the corresponding relationship between the adjacenl frames.And then the random sample consensus algorithm is used tn remove the outer points,then the obtained interior points are used to calcolate the adjacent pose of the mobile robot.The graph optimization method is applied to optimize the robot pose and globally consistent robot trajectory and the three-dimensionat model of the environment are obtained.(2)A loop closure detection algorithm baaed on convolution neural network for the mobile robot VSLAM is proposed,that is the loop closure delection algorithm based on Places365-VGG tor the mobile robot VSLAM A pretrained CNN model PIaces365-VGG is used in the loop closure detection algorithm of the mobile robot.The loop closure detection algorithm based on Places365-VGG of the mobile robot VSLAM first uses a pretrained CNN model Place365-VGG to extract the feature of the pool5 layer,and generate the feature vector of the image from the output of the layer,Then the cosine distance is used as the calculation method of the image similarity measure.The cosine distance between the two images is calculated,and the cosine distance is used to judge whether the two images are similar or not,so as to determine whether the closed-loop is detected.The lesults of the offline 3D VSLAM experiment based on standard dataset and the results of the online 3D VSLAM experiment in real scene show that the proposed mobile robot VSLAM system based on AKAZE tan effectively construct the three-dimensional map of the environment,accurately estimate the motion trajectory of the robot,and improve the positioning precision of the mobile robot.The validity and accuracy of the mobile robot VSLAM system bssed an AKAZE are verified.The loop closure detection results of the mobile robol VSLAM based on the standard dataset show that the proposed loop closure detection algorithm based on Places365-YGG for the mobile robot VSLAM can effectively detect the loops of the standard dataset.and the effectiveness of loop closure detection algorithm based on Places364-VGG for the mobile roobot VSLAM is verified.