Abstract: Auto-Focus technique is a main approach to hunt clear images which plays an important role in digital camera application. According to several unknown target under complicated condition, a novel auto-focus algorithm was proposed based on optical flow estimation. By calculating the optical flow of each input frame, the moving targets in scene image were tested as well as according to the moving characteristic, the interested real target was judged. Brenner sharpness evaluation method was improved. Meanwhile the evaluation function was established using two dimensions edge-gradient information. The response sensitivity of evaluation function was also increased via nonlinear-gain coefficient the impact of noise on evaluation value was decreased. Experimental results show that the proposed method can distinguish the interested main target in 50 times depths of field of different targets and evaluate the definition of varied images with random noise in 0.02 variance value effectively. And it is of a good ability of robustness for different images, Brenner function improves the peak stability margin 1 to 4 times by the algorithm, and it can be easily achieved on hardware.