A computationally efficient pyramid adaptive dynamic Hough transform (PADHT) is proposed to detect edges with arbitrary shapes and without prior knowledge. The binary pyramid structure is generated first. The edge map at the top pyramid level is obtained by using PADHT next. The edge map at each lower pyramid level is generated by PADHT also, but in a much smaller area determined by the edge map at the level immediately above. The analytic expressions of the arbitrary edge curves for the dynamic Hough transform are approximated by the rational Gaussian functions. The parameters of the Gaussian functions and the accumulated areas are changed adaptively according to the set of accumulated points. An energy function is used to determine which set of accumulated points represents the edge curve.