Image-based Indoor Positioning System Using Position Feature Detection

Location-based services have specific application scenarios in all aspects of life.Indoor positioning is an important part of location-based services and is closely related to people's life.Image feature take up a dominant part of the information that helps human and robotics understand the environment,and many image-based localization systems have been proposed.However,how to weigh the accuracy and cost of indoor positioning still requires research.An image localization system based on position feature detection is proposed.The system uses deep learning and local feature of image extraction algorithm to achieve positioning.In the offline state,the system collects the position feature of image and sets some manual marks at a suitable position.At the time,we measure coordinate of the points that indicate location information and establish the feature database.In the online state,the trained detection model is used to detect image features and obtain the feature points using image feature detection algorithm.From those feature points,we can get pixel coordinate of the feature points recorded in database by filtering,then matching image pixel coordinate and world coordinate of the feature points we calculate the position.When calculating positions,intrinsic parameters is needed.We provide camera calibration process which is simple and easy to perform,in addition,one camera only require one calibration process.In the training of the image detection model,the pre-trained model is used as the starting point,so the requirement for the amount of images is not large,and the amount of images to be stored is greatly reduced compared to the system which stores images of the whole indoor environment.The recognition process of the location feature points uses SIFT algorithm to extract the feature points of the image.The feature region in the image is relatively small.The use of the SIFT algorithm for the feature regions do not spend a lot of time.The coordinate points are matched by KNN algorithm and the RANSAC algorithm which are able to eliminate the error points to ensure their accuracy.The experimental results show that the above method can achieve higher precision positioning,and the positioning accuracy is higher than the existing methods.Moreover,the system does not require professional image acquisition equipment to collect data,and ordinary smartphones can complete image acquisition and positioning process.