Abstract: A new detecting algorithm based on gray-weighted kernel function was proposed to solve the proplem of low running rate and high false alarm within the moving target detection(MTD) in dynamic series of image. This algorithm firstly realized image sequence registration by using the biggest gradient block, then divided the image into 3232 sub-images. It could calculate gray-weighted kernel function for every sub-image and detect changing of gray-weighted kernel function by using Bhattacharyya coefficient as similarity principle for every sub-image. The moving target could be detected in sub-image which gray-weighted kernel function has changed. The testing result shows that the algorithm with batter performance of real-time and robustness can detect the moving target in real-time and suppress the influence due to image registration error and gray fluctuation effectively.