Comparing with the classical level set, the variational level set without re-initialization can avoid repeating initialization, which greatly reduces the algorithm's running time while using the edge gradient information of images to accurately capture the local structure.However, this model cannot adaptively obtain initial curve, and the model's topology cannot be changed to detect multiple targets.To solve the problems above, this paper proposes an adaptive contour based variational level set model for multiple target detection in complex background.First, the inter-frame difference algorithm is combined with K-means clustering algorithm to obtain multiple initialization curves, and then the noise is reduced by morphology method.This can estimate the position and the size of the moving target in complex background.The variational level set without re-initialization is further extended to multiple targets from single target, and the model's ability is improved to deal with the images of non-uniform gray.Experiments on standard database and real scene data sets indicate that the proposed method can accurately locate targets contours of different scales and gray to improve the evolution efficiency and accuracy of the algorithm.