Apply and Optimize 2D Object Detection in Assembling Components

In a world that human labor is becoming scarce while AI technology is fast developing, using Neural network to achieve 2D object detection to simulate workers is a frontier topic. In this paper, we try to discuss one spectacular utilize environment of applying 2D object detection, in which we want to use a camera to get picture input of a certain kind of component and detect it, then use robot hand to grab to and put it to somewhere else for next step of assembling. We are going to use YOLO v5 application and PyTorch neural network. We are going to show how to apply these algorithms and how we optimize them. During optimization, we are going to use label box editor as a tool for labeling. For conclusion, we will present the advantages and disadvantages of these new optimizations, by comparing them to normal method of object detection. We achieved some major advantages, including shorter running time and easiness of training. By contrast, we also meet some disadvantages, including losing accuracy and limited field of usage.