Abstract: For vehicle detection in complex environments, an improved Faster R-CNN algorithm is proposed. For the traditional Faster R-CNN algorithm that has problems such as missed detection, false detection, and slow running speed when applied to vehicle detection, this study adds more vehicle data sets in complex environments, improves the non-maximum suppression algorithm of the original model, and integrate batch norm algorithm to optimize it. Experiments have proved that the improved Faster R-CNN has a 4.86% increase in the comprehensive average recognition rate on the data set compared to Faster R-CNN, and the improvement effect is more obvious.