A Performance Review of 3D TOF Vision Systems in Comparison to Stereo Vision Systems

Abstract

The most common and well-known principle of 3D image acquisition is stereo vision (SV). This principle of 3D-image acquisition is already known and used for decades in the research community. The advantage of stereo vision to other range measuring devices such as laser scanners, acoustic or radar sensors is that it achieves high resolution and simultaneous acquisition of the entire range image without energy emission or moving parts. Still, the major disadvantage is the limited Field of View (FOV) and the correspondence problem. To enhance the FOV many techniques are researched such as rotating cameras (Kang et al., 1997; Benosman et al., 1996; Krishnan et al., 1996), increasing the number of the cameras (Kawanishi et al., 1998), and the use of a special optic (Liancheng & Feng, 2005; Lin & Bajcsyg, 2003). Also a combination of 1D-laser scanner and a SV system are proposed to overcome the FOV problem (Cheng et al., 2002). However, these systems are expensive and not easy too synchronize. Due to the correspondence problem algorithms need to be improved to give a lower percentage of false matches as well as better accuracy of depth estimates. Performance of algorithm needs to be evaluated over a broad range of image types in order to test their robustness (Dhond & Aggarwal, 1989). In the past years the modality of Time-of-Flight (TOF) imaging became more and more attractive to a growing research community (Schwarte_a et al., 1997; Schwarte_b et al., 1997). Because of the enormous progress in TOF-vision systems, nowadays 3D matrix cameras can be manufactured und be used for many application such as robotic, automotive, industrial, medical and multimedia applications. Due to the increasing demand of safety requirements in the automotive industry it can be assumed that the TOF-camera market will grow and the unit price of these systems in the mass production will drop down to ca. 100 € (Hussmann & Hess, 2006). For all application areas new accurate and fast algorithms for 3D object recognition and classification are needed. As now commercial 3D-TOF cameras are available at a reasonable price the number of research projects is expected to increase significantly. One early example of using a TOF-camera based on the Photonic-Mixer-Devices (PMD)-Technology for 3D object recognition in TOF data sets are presented in (Hess et al., 2003). In this paper the transition from a general model of the system to specific applications such as intelligent O pe n A cc es s D at ab as e w w w .ite ch on lin e. co m