To detect a moving point target that is submerged in noisy background is a very difficult problem.A new approach to moving target detection based on multiscale geometric technique is proposed in this paper.A 4-D scale spatio-temporal space is constructed by multiscale approach using Gaussian as a filtering kernel.A motion constraint equation for moving target is derived by the first order geometric features.A technique for solving the problem of motion detection and velocity estimation based on the velocity identity and pixel connectivity of the target image is described.A coarse-fine search algorithm of the clustering point or area for constrainted straight lines on the velocity plane is outlined.An accuracy measure for precisely how to search an optimal scale of Gaussian is given.The experimant and comparison results are presented to support the research.