Abstract
In the artificial environment like inside of a building, the scene from the robot's view point usually consists of line segments. This paper proposes a new effective algorithm to extract line segments from a picture for a robot vision. It should be paid attention to the fact that directions of gradient of brightness on a line segment are almost equal in a picture. In this algorithm, pixels having the following properties are firstly connected together to generate a small area: (1) Magnitude of the gradient of brightness is greater than a threshold value. (2) Direction of the gradient of the pixel falls within a given small range of directions. (3) All pixels are connected each other. Such a small area is generated like a stick on the border line of an object in the original picture. Line segments can be extracted by applying least square method to each generated sticklike areas. The advantages of this method are; (1) the algorithm needs only one scan to generate all sticklike small areas and is suitable to process an original picture in short time, (2) the effect of local noise is greatly reduced with statistical calculations to fit line segments on the small areas, (3) sticklikeness of each small area can be evaluated by ellipticity of its equivalent ellipse. This paper also shows experimental results to prove that line segments can be effectively extracted from an original picture by uniform algorithm on the proposed method.
