INTRODUCTION - Video Mosaicing
Separation Bar

The subsea medium constitutes a challenging environment for image acquisition and processing (Garcia et al. 2002c). When compared with land and aerial applications, the light in the underwater medium is subjected to intense scattering and absorption (Garcia et al. 2005), which severely limits underwater imagery in terms of contrast, sharpness and range (Neumann et al. 2004, Garcia et al. 2003c). Often, the only possibility to obtain a useful visual representation of the seafloor is by composing a large number of close-range images (Gracias 2000, Negahdaripour 2001). Image mosaicing deals with the process of combining the information from multiple images of the same area, to create a single representation with extended field of view (Irani et al. 1995, Garcia 2001d). In order to construct ocean floor photo-mosaics, the individual images forming the mosaic are usually obtained by setting a camera on an underwater vehicle (Negahdaripour et al. 1998). The camera looks down and the acquired images cover a small area of the ocean floor. The automatic detection of a number of points in one image, and its correspondences in another, allows the estimation of the motion between them, and the computation of the motion between them (Negahdaripour et al. 2006). A number of techniques for seafloor mosaicing can be found in the literature. Once of the most relevant groups is the Instituto Superior de Robotica in Lisbon (Gracias et al.1998; Gracias et al. 2000, Gracias et al. 2001), who presented different strategies based on the motion detection of consecutive frames. We also can find works focused on the error reduction of the accumulated drift, such as the work carried out by the researchers from Stanford University, in conjunction with Monterrey Bay Aquarium (Marks et al., 1994; Marks et al. 1995; Fleischer et al. 1996). Probably the most relevant work is the one carried out by Negahdaripour et al. at the Univ. of Miami, where dense optical flow techniques were applied, enabling the 3D motion estimation from the spatio-temporal gradient of the images (Xu et al 1997, Negahdaripour et al 1998a; Negahdaripour et al. 1998b; Negahdaripour et al 1999; Xu et al., 1999, Negahdaripour et al., 2002). On the other hand, Rzhanov et al. at the Univ. of New Hampshire proposed a system based on the motion detection in the frequency domain by using the Fourier transform (Rzhanov et al., 2000). Also, we should mention the work carried out by the researchers of the Heriot-Watt University (Trucco et al., 2000a; Trucco et al., 2000b; Lots et al., 2001), who developed techniques able to reduce de drift error based on a set of features tracked in the image sequence.

Figure 1. Sample mosaic of the Lucky Strike area in the mid-Atlantic Ridge (Escartin et al. 2006). The mosaic covers an area of more than one square kilometer, while keeping the high resolution of the individual images. Images provided by Dan Fornari (WHOI).

More recently, Singh et al. from WHOI have successfully implemented a mosaicing system based on a careful detection of features and a Simultaneous Localization and Mapping (SLAM) approach (Eustice et al. 2005, Roman 2005). Finally, the VICOROB group of the University of Girona has also proposed different methods in order to improve the detection of corresponding points in underwater images by using texture features (Garcia et al. 2001b, Ila et al. 2004), and also has proposed different methods aiming the automatic and robust reconstruction of photo-mosaics (Garcia et al. 2000, Garcia et al. 2001a, Garcia et al., 2001c, Garcia et al. 2003a). In another work (Garcia et al. 2002b, Ila et al. 2004, Garcia et al. 2005), Garcia et al. presented a new methodology with the goal of integrate the mosaics with the map building and the automatic robot localization (SLAM) (Ribas et al. 2006, Estrada et al. 2005, Martinelli et al. 2005, Folkesson 2005). Moreover, our group has already proposed solutions to merging control and sensing, closing the loop of the control architecture of the robot by using photo-mosaics (Garcia et al. 2003b, Carreras et al. 2006). A sample mosaic of a large area of the Mid-Atlantic ridge is illustrated in Figure 1 (Escartin et al. 2006).

Fortunately, technology is improving with the advent of new generation of sensors able to provide data with high visual quality, differently from only few years ago. This is the case of the acoustic cameras, such as the Echoscope or DIDSON. In the literature, one can find nowadays few works to study solutions for the data fusion coming from optical and acoustical cameras. Therefore, the effort is concentrated in the recovery of detailed 3D information of the underwater floor. To the best of our knowledge, only the Underwater Vision and Imaging Laboratory of the University of Miami (Sekkati and Negahdaripour 2006b, Negahdaripour 2005a, Negahdaripour 2005b, Negahdaripour 2006) and the Department of Computer Science of the University of Verona (Castellani et al. 2004, Castellani et al. 2005) have started to investigate this promising emerging area of underwater 3D mosaics reconstruction.

<< Next >>