An Integrated IMU, GNSS and Image Recognition Sensor for Pedestrian Navigation

Abstract

Foot mounted inertial sensors provide a promising method for accurate pedestrian positioning. By mounting sensors on a user’s foot, the accumulated position drift from dead reckoning can be reduced by applying zero velocity updates using a Kalman filter every time a user takes a step. However, such a system will still suffer from position drift unless occasional position updates are available. This paper describes a novel method for restricting such position drift using an image recognition algorithm. Firstly, a database of images and their locations is constructed over an area of interest. A user then navigates the area using foot-mounted inertial sensors and a video camera. As images are acquired, they are used to search the database of images using the Image Bag-of-Words algorithm. When new images are successfully matched with images in the database, the position from the database is used to update the inertial position using a Kalman filter. GNSS updates can also be used in the filter when available. The integrated inertial and vision system is demonstrated to provide better than 10m accuracy (typically 1-5m) over a period of 21 minutes. The system is relatively inexpensive, could run in real-time, does not require costly infrastructure, and could be deployed over larger areas.