Research On Binnocular Indoor Positioning Algorithm Based On Natural Scene

In recent years,with the development of artificial intelligence,a variety of smart home products can be seen in many homes.And people's desire for further intellectualization of home life is becoming more and more intense.More and more people are focusing on home service robots and they expect the home service robots can serve our life more intelligent.If so,the home service robots will bring a new life change.Due to technical difficulties,service robots have not yet been widely used.Therefore,the research and development of intelligent home service robots have become a hot spot in current research,and at the same time it has promoted the development of the entire robot industry.Due to the complex indoor environment and the weak GPS(Global Positioning System)signal,positioning navigation has become a major problem for home service robots.At present,there are endless researches on this topic,among which the wireless signal positioning represented by WLAN(Wireless LAN)is easy to attenuate and the positioning accuracy is low;active vision positioning represented by laser radar positioning is greatly affected by the environment.,and the cost is higher;the positioning based on passive vision has great challenges in real-time.Among the multiple positioning methods,the passive vision that simulates the human eye system for scene location is most in line with the development direction of artificial intelligence.The collected visual images have higher information volume and can be used for later target tracking and recognition.This paper presents a solution to the problem of indoor positioning of home service robots based on binocular feature extraction and calibration of natural scenes.The current method was compared in terms of real-time performance and accuracy.The steps of indoor positioning were integrated and an improved algorithm was proposed.The main work accomplished by this article includes:(1)Introduce the research background and significance of home service robot positioning system and research status at home and abroad.Identify the current robot positioning process and analyze the difficulties.It draws out the binocular positioning and its research significance,and summarizes the development of its key algorithms.(2)Analyze the binocular indoor positioning process as a whole,introduce thepositioning of major modules and analyze the key technologies of each module.Aiming at the inaccuracy of the calibration parameters caused by manual operation in the commonly used calibration algorithms,several calibration calibration camera parameters are proposed.The calibration results are analyzed using a variety of calibration parameter measurement methods to obtain the optimal camera parameters.Finally,the experiment verifies the availability of the calibration method and calibration results.(3)Compare the current several feature extraction and matching algorithms,and combine the feature extraction time and recovery quantity to determine the robustness of the feature.Aiming at the high time-consuming problem of the feature extraction and matching algorithm used in the current research,a stereo matching algorithm that compensates the ORB matching features is proposed.The ORB feature is used to reduce the time consumption of feature extraction.The KNN algorithm reduces the time consumption of stereo matching and uses the compensation feature.The way to make up for the status of ORB's less available features in matching.Through experiments,it is proved that the number of ORB features after compensation is significantly improved,and the time consumption is also significantly improved compared to other feature algorithms.(4)In the online positioning module,the calibration scene identification uses an image recognition algorithm based on feature matching,and the number of intra-area point features filtered by RANSAC is taken as the similarity degree between the current natural scene and the calibrated natural scene.This paper proposes a method of partitioned iterative matching for natural scene recognition by increasing the number of intra-site points and the time-consuming increase of algorithm iterations caused by the low proportion of intra-random points in RANSAC.Strictly limiting the number of iterations to improve the efficiency of the algorithm,and experimentally proved the feasibility of this method,and greatly reduce the scene recognition time.(5)By modeling the laboratory scene and extracting scenes that cannot be integrated into the scene by manual landmarks,extract offline scenes with relatively centralized scene features in the same plane.Finally,the whole solution was verified and the final database sampling proposal was given.