Based on identified driving lane, identification will be extended from single-lane to multi-lane by way of ROI (region of interest) control. The concept of confidence was introduced in order to judge the degree of accuracy on each lane, and the calculation of confidence was given under the complex condition. If the accuracy of some lanes was below the threshold, those ROI of the search parameters would be adjusted accordingly so as to avoid the search target being out of the established ROI. At the same time, by analyzing the image sequences, with the combination of the structural characteristics of each lane, the identification result was judged, that could effectively eliminate the effect of vehicles driving on adjacent lanes and remain no lane next to the driving lane upon identification. In this way, the recognition reliability increased. It is proved that this method could improve the accuracy of multi-lane identification.