As we all know, some bad driving behaviors such as using mobile phone, eating, drinking etc., may lead to traffic accidents. In order to reduce these potential dangers, we proposed a target detection network to detect the abovementioned dangerous driving gestures. We have improved the YOLOv4 target detection network and designed the DDGNet- YOLO target detection network, so as to reduce the detection time of dangerous driving gesture. Our algorithm have been performed on the CVRR-HANDS 3D dataset. The mAP@0.75 of dangerous driving gesture is 76.4%, and the average real-time processing speed of the algorithm was 143 frames per second.