Recently, to appropriate the demand of diverse open traffic environment, autonomous driving technology is rapidly advancing. There has been also increasing interest in the optimization of object detection technology in terms of precision, speed, generalizability as well as software and hardware requirements. However, it is difficult to balance network parameters between performance of detectors. Considering the limited computational resources for autonomous vehicle. In this paper, we propose an efficient and light-weight object detection algorithm based on YOLOv5 improved by Knowledge Distillation. We pre-train a high-performing teacher model on the KITTI Dataset to transfer the knowledge to a less performant small student model. Multi-Teacher Distillation is applied on the CARLA datasets of different maps to improve the generalization ability of the student model. Finally, we obtain a high-performance D-YOLO-C detector. Experiments show that our method achieves to increase the mAP score to 73.8% on KITTI + CARLA dataset with little sacrifice to inference speed and computational resources.