Abstract
This paper proposes a real-time image processing system for wheelchair recognition at elevator lobby. This system extracts objects with frame difference method. From the image of extracted objects, heads of people are detected by the Hough transform. Vertical shooting from ceiling improves the performance of people counting since the detected head has high roundness and occlusion of people is prevented. Two feature quantities are introduced to recognize wheelchairs. They are the area value and the ratio of traveling length to vertical length. Since these quantities require a method for drawing the contour line. This paper proposes a new simple method to draw a contour line. The effectiveness of this system is confirmed through experiments.
