Research On Front End Key Technologies Of Mobile Robot Vision SLAM In Indoor Environment

In recent years,with the development and maturity of the robot industry,mobile robots with high autonomy and flexibility have become popular research directions in various fields.Instant positioning and map construction technology is a key technology for mobile robots to achieve working and autonomous walking.The core part of SLAM refers to the process in which the robot uses only the information collected by its own sensor to realize pose estimation and incremental map construction in an unknown environment.Among them,the visual SLAM is different from the traditional SLAM technology using laser and sonar.It has the advantages of low price and strong environmental adaptability,and has been widely used in various fields of SLAM technology.However,visual SLAM relies on image information for incremental map construction,and it is necessary to rely on the front end to collect and process image information,thereby obtaining an image capable of carrying complete scene information,that is,a key frame.Whether the key frame can be quickly and accurately selected will directly affect the accuracy and reliability of the visual SLAM.Therefore,the paper uses neural network to construct the key frame adaptive selection model,and utilizes the powerful optimization and data processing capabilities of the neural network to overcome the shortcomings of the traditional SLAM key frame selection algorithm,such as slow speed and low precision,and has obtained good experimental results.This proves that the method has certain practical significance.In this paper,the traditional BP neural network and the PIO-BP neural network are used to construct the key frame adaptive selection model.Firstly,the ROS robot is used as the carrier,and a large number of experiments is used to select the parameters that can affect the number of key frames selected,that is,the linear velocity and angular velocity of the robot movement,and they are used as the input of the BP neural network model.The key frame filtering step size is selected as the output of the network model,and the nonlinear relationship between input and output is trained.The key frame adaptive selection model based on BP neural network is obtained.Secondly,aiming at the slow convergence speed of traditional BP neural network,the pigeon population algorithm is used to optimize the network parameters of BP network,which improves the convergence speed of the network.At the same time,the related improvement strategy is proposed for the problem that the pigeon group algorithm is easy to fall into the local extremum.The weight of the individual speed update in the algorithm is increased,and the optimization ability of the pigeon group algorithm is optimized,witch significantly improved the performance of the keyframe adaptive selection model.Finally,three kinds of feature detection and matching algorithms are compared and the ORB feature detection algorithm is used to detect and match the image features.The experimental results show that the improved key frame adaptive selection algorithm effectively enhances the real-time and accuracy of SLAM,and has certain practical value.