Vehicle and pedestrian detection is significant in autonomous driving. It provides information for path planning, lane selection, pedestrian and vehicle tracking, pedestrian behavior prediction, etc. In recent years, the state-of-the-art object detection algorithms have been emerged on the base of deep convolutional neural networks, which can get higher accuracy and efficiency detection results than traditional vision detection algorithms. In this paper, we first introduce and summarize some state-of-the-date object detection algorithms based of deep convolutional neural networks and the improvement ideas of these algorithms. Their frameworks are extracted. Then, we choose several different algorithms and analyze their running results on challenging datasets, Pascal VOC and KITTI. Next, we analyze the current detection challenges as well as their solutions. Finally, we provide insights into use in autonomous driving, such as vehicle and pedestrian detection and driving control.