Online map building based on ultrasonic and image sensor

Online building of a local map is important for robot navigation. While an ultrasonic sensor is fast at acquisition of the environment model, it is poor at angle resolution and error-prone as false reflection occurs. On the other hand, while an image sensor is slow at image processing, and suffers from the lighting condition in an indoor environment, it is good in resolution. In this paper, we try to merge these two types of sensor information to produce an occupancy grid representation of the environment which is good for extraction of floor connectivity. In particular, we propose the principle of the shortest nearby range values to address the false reflection issue. Then we use k-mean clustering to classify the floor color. A safety index representation is used as a basis for fusion of the echo and image map. Experiments are done which show the usefulness of the fusion map.