Sidewalk Border Detection using Stereo Images for

Abstract

We propose an algorithm to extract sidewalk regions and recognize components of sidewalk border, width of sidewalk, and existence of bottleneck points and opening points from outdoor road images obtained from a stereo vision camera mounted on a low-speed vehicle. The elevation map is constructed by removing noises from the inputted stereo images. The probability that each point belongs to the sidewalk border is calculated from the map, and the closed loop that is considered to be corresponding to the sidewalk border is extracted. Then, the type (such as curb and wall) of border component is determined. The sidewalk attributes such as width, components of borderline, presence of bottleneck point, and opening point are obtained by integrating these results among a series of inputted images. We applied this algorithm to the images obtained from a low-speed, outdoor vehicle (based on a commercial electric scooter) and calculated its efficiency.