Real-time Obstacle Map Reconstruction by Sensors with Limited Field of View for Vision Based Autonomous Robot

Sensors on mobile robot usually have limited field of view which is unfavorable for obstacle detection. To make indoor mobile robot avoid obstacle automatically using these sensors, this paper presents a method to reconstruct obstacle map originated in human behavior. It gives every observed obstacle a credibility and if it is not observed, the credibility will decrease in time until 0. By this way, the different observed obstacles information with time can be integrated into one obstacle map. It does not use feature mapping and only does some matrix operations on the obstacle information so it has high reliability. The experiment shows the algorithm has good performance and can fit in variable environment.