Hybrid Dual Camera Vision Systems

In this paper we describe the Hybrid Omnidirectional Pin-hole Sensor (HOPS) dual camera system. Since its joint camera calibration leads to a fully calibrated hybrid stereo pair from which 3D information can be extracted, HOPS suits several kinds of applications. For example, it can be used for surveillance and robot self-localization or obstacle detection, offering the possibility to integrate stereo sensing with peripheral/foveal active vision strategies: once objects or regions of interest are localized on the wide-range sensor, the traditional camera can be used to enhance the resolution with which these areas can be analyzed. Tracking of multiple objects/people relying on high-resolution images for recognition and access control or estimating velocity, dimensions and trajectories are some examples of surveillance tasks for which HOPS is suitable. Accurate obstacle detection, landmark localization, robust ego-motion estimation or three-dimensional environment reconstruction are other examples of possible applications related to (autonomous/holonomous) robot navigation in semi-structured or completely unstructured environments. Some preliminary experiments performed to solve both surveillance and robot navigation with encouraging results are discussed through the paper.