3-D object modeling from occluding contours in opti-acoustic stereo images

Utilizing in situ measurements to build 3-D volumetric object models under variety of turbidity conditions is highly desirable for marine sciences. Feature-based structure from motion and stereo methods can become ineffective under poor visibility, where matches cannot be reliably identified in low contrast optical images. To address this, a solution is proposed by utilizing a multi-modal stereo imaging technique with coincident optical and sonar cameras. Moreover, the challenges of establishing the complex opti-acoustic correspondences are avoided, by employing 2-D occluding contours that are the images of 3-D object rims as structural features. As we circle our target object, we process matching 2-D apparent contours to construct the corresponding 3-D object rim, and compute the trajectory of the stereo rig by optiacoustic bundle adjustment in order to transform the 3-D object rims into registered samples of the object surface in a reference coordinate system. In addition to exploiting range measurements from sonar that offer unique advantages, the proposed paradigm enables computing both the 3-D positions and local surface normals of 3-D contours, leading to improved object reconstruction accuracy. We demonstrate the performance of our method based on the 3-D surface rendering of certain objects, imaged by an underwater opti-acoustic stereo system.