Research On SLAM And Target Detection Technology Of Mobile Robot Based On Monocular Vision

With the rapid development of emerging technologies such as artificial intelligence,computer vision and big data,mobile robot technology and target detection technology have greatly promoted the entry of people's lives,making people's lives move towards an intelligent society.Mobile robot SLAM technology and target detection technology,as the core key technologies to realize a fully intelligent mobile robot,will provide important technical support in the fields of mobile security,robot service,express delivery and security inspection,and have broad market prospects in the future.This paper studies the fusion of vision and IMU information in the monocular visual inertial navigation SLAM technology,as well as the key technology of the positioning system and the complex background,occlusion and small targets in the human detection technology.Not only does it design an environment in which features are lacking The monocular visual inertial navigation SLAM system with both robustness and robustness has also been designed with a new convolution module and basic network,which makes the detection network of this paper significantly improve the detection performance of the human body and human head.The main research work and innovation of this article are as follows:(1)The principle and model of the camera and IMU sensor of the mobile robot SLAM system were analyzed and deduced,a monocular visual inertial navigation SLAM system that can work in the environment with lack of features was built,and each of its modules In detail,the bag-of-words library model of the loop detection module is improved,and a dedicated data set is produced.The stability and robustness of the monocular visual inertial navigation SLAM system designed in this paper are verified by the dedicated data set.(2)The related methods and basic knowledge in target detection technology are analyzed and elaborated,and the R-CNN and YOLO series algorithms are comprehensively analyzed from the technical means,network structure and algorithm flow,and Experimental verification and comparison of the effect and performance are carried out,which provides a sufficient basis for the subsequent selection and improvement of the human body and head detection network.(3)The detection scheme for the detection method of human body and head is proposed,the exclusive data set of this article is made,and a new type of convolution module is proposed.The basic network structure is designed based on the new type of convolution module.In order to maintain the network The high-resolution information of the output features is designed for the network output mode,and the pre-training weight experiment is designed,which proves that the feature-rich pre-training can effectively improve the generalization performance of the network;the data expansion experiment proves the color jitter and data stylization strategy The detection effect has been significantly improved;the model experiment proves that the proposed IIHNet network has excellent detection effect;the detection effect and inference experiment prove that the IIHNet network not only has excellent detection performance,but also its inference performance has been significantly improved.