Study Of Dense Depth Acquisition Techniques Based On Structured Light And Stereo Matching

The goal of image depth acquisition technology is to reconstruct the 3-D geometric information of objects in the scene from one or a sequence of images. Recently, it has been widely applied in the fields of factory automation, virtual reality, computer aided design (CAD), entertainment computing and reverse engineering, etc. Stereo matching is a classic depth acquisition technology. It matches the correspondence of pixels or features in multiple photos of the same scene which are taken from different positions, and thus the depth information of the 3-D scene can be reconstructed from 2-D images. Because of its practicability, effectiveness and high automatization, nowadays, it is one of the hottest research topics in computer vision and computer graphics.Depth acquisition technologies based on stereo matching has two main categories, namely, active stereo matching and passive stereo matching. In active stereo matching technologies, illumination information is projected into the scene, such as laser scan lines and structured light, in order to increase the identifiable features in the scene, and to minimize the difficulties involved in correspondence. On the other hand, passive stereo matching does not add any assistant information into the scene. In this paper, taking into account the advantages of both active and passive stereo matching, we propose a dense depth acquisition approach based on 2-D pseudo-random structured light and dense image stereo matching. It requires taking only one image of the scene, and is able to acquire depth for every pixel in the image. Moreover, it is capable of dealing with scenes which are lack of texture.There are mainly three steps in our approach. Firstly, a 2-D binary pseudo-random structured light pattern is generated and cast into the scene by an off-the-shelf projector. Secondly, the image of the scene under structured light illumination is rectified to remove radical distortion, and the matching information of corresponding pixels is obtained via dynamic programming stereo matching algorithm. Finally, according to the geometric parameters of the camera and the projector obtained by calibration, we acquire the dense depth by applying triangulation based on least squares. Meanwhile, based on the traditional dynamic programming stereo matching algorithm, we add vertical constraints into the computation of the optimal matching path, i.e., constraining the current optimal path by the previous ones. Thus, the scan-line effect of the depth map is eliminated, and in accuracy, the improved matching algorithm outperforms most of the state-of-the-art binocular stereo matching algorithms with similar efficiency on Middlebury College's stereo web site.In our approach, only a projector and a digital camera are used to acquire dense depth information with little manual intervention. Experiments show that high quality dense depth can be obtained using our approach, and the accuracy is reasonable considering the relatively low hardware requirement and wide applications.