Object location is a principal problem in the field of machine vision. In order to eliminate the negative effects of noises, an improved Hausdorff distance based on edge-strength was proposed for object location. Edge length, edge gradient and edge closure were used for edge weights computation, and a location function was then established by adopting weighted Hausdorff distance. Distribution estimation algorithm concerning related multi-variables was used to avoid local minimum during optimization, which described the relationship between matching parameters with a probability model. The tasks of location were realized by probability model evolution. Experiments show that the algorithm outperforms other approaches in both accuracy and rapidity.