Image Analysis And Indoor Obstacle Location Based On Dual Camera

In the field of image processing, obstacle recognition has always been an important and difficult problem. Because of the limitations of the information provided by the image, until now there is no way to accurately and effectively recognize all types of obstacles. Although binocular camera provides the important information of depth which is beneficial to realize image recognition, it costs too much which restricts its application. While, it is too difficult to locate the obstacle by low cost ordinary camera. Since the beginning of the seventies of the 20 th century, even though the field of image segmentation, as an important part of obstacle recognition, evolves numerous image segmentation algorithms, no conventional theory can accurately and effectively segment all types of obstacles without artificial participation.This paper is dedicated to the research on the effectively segmenting and locating obstacles without artificial participation. However, it is impossible to find a method to identify all types of obstacles accurately. In this paper, based on the application background of unmanned wheelchair, i.e. indoor environment, the model of obstacles can be simplified.First, we use two ordinary webcams to acquire original images. However, these images can't be used to recognize obstacles. Therefore, we calibrate them to obtain undeformed images. Second, in order to obtain the position of obstacles from images, we use the edge detection method and image segmentation method to locate them. Then, through convex hull extraction, the key points of obstacles in images are extracted. At last, by using the known obstacle model and the conversion relation between the image coordinate system and the spatial coordinate system, the location of obstacles in the room can be determined.