An indoor visual positioning system using points of interest detection for mobile robot application

Positioning of a mobile robot becomes significantly important in robotic navigation applications as it affects the subsequent motions and path planning operations of a mobile robot. Among those technologies employed in mobile robot positioning, visual positioning has aroused many interests as it does not require the installation of the additional infrastructure. However, its computation complexity is high which prevents it from the implementation in mobile robots. In this paper, a novel method is introduced in which convolutional neural network (CNN) based techniques and digital maps that are used to store the point-of-interest (POI) locations are employed to quickly determine the location of a mobile robot. The method first trained a CNN network to recognize objects which are identified as the POI objects and they are then tagged with their corresponding locations in a digital map. During the positioning phase, the POI objects are recognized together with their positions and the positions of the mobile robot can be determined using a geometric metric method. The positioning system has been implemented and tested in an indoor environment, and the results show that the approach is very promising.