Research On Indoor Target Detection Method Used In Intelligent Blind Guide System

In recent years,the key technology in the field of artificial intelligence-computer vision technology has made great progress,and many brilliant technologies have appeared in multiple directions,such as object detection,semantic segmentation,instance segmentation and object recognition.And the landing and application of computer vision technology are also one of important research directions.Nowadays,various computer vision technologies have spread around people,such as face payment,pedestrian detection,text recognition,and applications on robot and drone platforms.In the field of blinding technology,computer vision technology can exactly fill the blindness of the blind.After 2010,the emergence of RGB-D sensors added depth information to color images.This paper mainly studies the theory and implementation of indoor object detection and location combining RGB-D technology and target detection technology.To begin with,this paper studies the construction and implementation of an indoor object detection platform: the first step is to build a wearable terminal,which is used to collect image and voice data and perform voice interaction with the user;the second step is to build a server and run target detection on the server Related procedures,and realize the detection and prompting of forward obstacles and object search and orientation prompting;the last step is to establish communication between the server and the terminal,and in order to ensure the synchronization of the communication,multiple sign characters are also set.Secondly,the method of target ranging and angle measurement based on depth image is studied.Aiming at the phenomenon of voids in the original depth map,this paper uses a depth map repair algorithm that combines color images and depth images;in order to refine the repaired depth image,a guided filter is used to filter the depth image.In order to obtain the true distance and azimuth of the target and the user,the ranging angle measurement method based on the depth image is studied,and the ranging error is corrected.Finally,this paper studies the indoor target detection algorithm based on neural network.In this paper,Efficient Det is used as the target detector,and Efficient Det-D3 is used as a suitable model by comparing the accuracy,speed and calculation of Efficient Det-D0?D7.In order to obtain the depth value of the target after the target detection is completed,this paper proposes a method based on superpixel clustering: first extract the target from the target detection result;then perform superpixel segmentation on the extracted color image;and then use the superpixel block For sample points,the average depth of the superpixel block is the sample value,and density clustering is performed;finally,this paper proposes a strategy to judge the category of the target and calculate the target depth value.