Stereo Vision Matching using Characteristics Vectors

Stereo vision is a usual method to obtain depth information from images. The problems encountered when applying the majority of well established algorithms to provide this information are due to the high computational load required. This occurs in both the block matching and graphical cues (such as edges) matching. In this article we address this issue by performing an image analysis which considers each pixel only once, thus enhancing the efficiency of the image processing. Additionally, when matching is carried out over statistical descriptors of the image regions, commonly referred to as characteristic vectors, whose number of these vectors is, by definition, lower than the possible block matching possibilities, the algorithm achieves an improved level of performance. In this paper we present a new algorithm which has been specifically designed to solve the commonly observed problems which arise from other well know techniques. This algorithm was designed using a previous work carried out by the authors in this area to determine the descriptors extraction processes. The complete analysis has been carried out over gray scale images. The results obtained from both real and synthetic images are presented in terms of matching quality and time consumption and compared to other published results. Finally, a discussion is provided on additional features related to the matching process.

Related material