Neural networks for 3-D motion detection from a sequence of image frames

A three-dimensional (3-D) motion detection system is described. This system consists of three stages; the rough motion detection stage, the moving object extraction stage, and the 3-D motion detection stage. Five neural networks, the correlation network, the rough motion detection network, the edge enhancement network, the background remover, and the normalization network, are designed for the implementation of these three stages. The processing time of the system depends on the number of moving objects detected. Normally, it will take about 9s on a sequential machine, such as the 80386. Two moving objects and their motions were detected correctly with this system.<
>