Digital video camera has the properties of large noise, uncertain location of main body and high requirement of real-time. The disadvantages of traditional auto-focusing algorithms were discussed in detail and a new auto-focusing algorithm was proposed. A novel focus measure filter called First Derivative of Gaussian (FDOG) was developed to estimate the degree of focusing correctly by the suppressing high frequency noise. To enable the auto-focusing system to focus on the exact main body, an automatic region-of-interest selection strategy based on the principle of nearest-subject-priority was adopted. Furthermore, combined with mountain-climbing search, binary search and quadratic interpolation, the proposed real-time search strategy greatly reduces the convergence time by about 40%, which enables the video camera to fast focus. The proposed algorithm is successfully implemented on a prototype digital video camera and experiment results are given to demonstrate the advantages of the proposed auto-focusing algorithm.