2D occupancy-grid SLAM of structured indoor environments using a single camera

Occupancy-grid simultaneous localisation and mapping (SLAM) has traditionally been implemented using range sensors such as lasers and sonars. This paper presents an implementation of occupancy grid SLAM using a single camera as the only on-board exteroceptive sensor and in a manner that is similar to a laser-based system. Range information is extracted from each image using a new approach. First, the ground plane is determined based on the orientation and height of segmented regions, then virtual rays are cast into the camera’s field of view from the optical centre to the intersection of each ray’s 2D projection with the ground boundaries. Occupancy-grid SLAM then uses the extracted depth information to build a dense map of the robot’s setting. Experiments are conducted inside a real lab setting and results prove the success of our system.