An Indoor Position System Based On Object Recognition By Using Panoramic Image

Indoor positioning refers to the real-time position monitoring of people and objects in the indoor environment through a high-precision indoor positioning system.At present,the indoor positioning system has been widely used in indoor unmanned aerial vehicle(UAV),goods sorting car in the indoor warehouse environment,personnel and goods management and other scenarios.In recent years,with the development of computer vision,the research of indoor positioning system based on vision has attracted extensive attention in industry and academia.This paper presents an indoor warehouse system based on a new panoramic camera.The panoramic camera creates a 360-degree panorama by reflecting external incident light onto an image sensor at the center of the bottom of the camera through a conical mirror.The image is transmitted to the integrated chip to output a panorama image,and the sensor is located in the center of the panorama image.The proposed work exploits the following facts:First,we design custom tags as scene feature control points and build warehouse scenes for indoor positioning.We use panoramic cameras to collect images of warehouse scenes in different positions to build a database for object detection.We build a single tag acquisition device and we construct a database for custom tag retrieval.Second,considering the defects such as severe distortion and large scaling variation of the objects in panoramic image,the object detection algorithm based on deep learning is designed to realize the detection of custom tags in indoor warehouse scenes.Third,based on the classical convolutional neural network,a feature extractor is constructed to extract features from a single custom tag and build a feature retrieval library.The content information of the custom tag in the indoor warehouse scene was determined by comparing it with the feature retrieval library.Fourth,precise positioning can be achieved by the position information of custom tags and angle relationship.In fact,the image retrieval stage is the inheritance of the target detection stage,and the precise positioning stage is also affected by the target detection and image retrieval.Finally,the positioning system proposed in this paper can achieve high accuracy,high speed and strong robustness,which can meet the requirements of indoor warehouse positioning.Sets of contrast experiments are designed to verify the performance advantages of using Faster R-CNN embedded FPN and Deformable Convolutions in object detection stage and using Res Ne Xt101 as benchmark network in image retrieval stage.