Abstract: Aiming at the safety problems of vehicle collision and scraping in grain dump, an intelligent grain vehicle AEB system based on vision and millimeter wave radar was designed.The system uses the single-stage deep convolutional neural network YOLOv5s to detect objects in the image, and uses the threshold screening method and the life cycle characteristics of millimeter wave radar false alarm to filter the millimeter wave radar data to sense obstacles.The vehicle is braking by using the hierarchical braking strategy.The real vehicle experiment can effectively realize the anti-collision function.Experimental results show that the anti-collision function and visualization effect of the system are better than that of a single sensor, and the system has real-time performance, effectiveness and safety.