Abstract: A new generalized hough transform algorithm based on importance sampling framework is proposed to solve the problem of large computation and low efficiency of traditional generalized Hough transform algorithm in object detection and the problem of false detection based on feature matching algorithm for zoom and rotate images.The generalized Hough transform is placed in the importance sampling frame, the edge points are filtered according to the significance measurement value, and are taken as the voting elements of the generalized hough transform.And the weighted multi-resolution analysis method is adopted to greatly reduce the computation.Through a variety of experiments (shrinkage, rotation, noise, shielding, etc.) and comparison with other similar methods, it is verified from the theoretical basis and concrete implementation that the method has higher efficiency and better accuracy, and can meet the real-time and high precision requirements in industrial detection.