A neural network algorithm for longitudinal motion stereo

Summary form only given. An algorithm for longitudinal motion stereo based on a discrete parallel neural network is discussed. Longitudinal motion stereo is a method to infer depth information from a forward or backward moving camera. Existing algorithms have some problems associated with the location of the focus of expansion (FOE), camera orientation, and surface orientation. The present algorithm allows the camera to move along its optical axis freely, needs no information on the FOE, and makes no requirements on the surface orientation. The algorithm uses a Gabor correlation operator to extract image features and employs a discrete parallel neural network to compute the disparity field based on the Gabor features. A depth map is then derived from the disparity field by simple algebraic computations