Object Detection Method Based on CNN and Camera Calibration

As an important storage device, the dense cabinet system has the advantages of compact structure, low space occupation, and large storage capacity. It has been widely used in the warehousing industry. However, due to the unstable opening position and the obstruction caused by the high cabinet body, it is difficult to perform real-time video monitoring of the opening position. In this paper, we propose a CNN method for detecting the opening position of dense cabinets based on image data. The improved ShuffleNet network is used to extract image features, and an edge point loss function is introduced to train the network. After the proposed method, the precise pixel coordinates of the centerline position at the opening of the dense cabinet are obtained, and the camera calibration method is used to transform them into real distance values, guiding the mobile monitoring camera to reach the opening position and achieve real-time video monitoring. The application result in a real dense cabinet system shows that our method can accurately detect the opening position of dense cabinets and has the foundation for engineering applications.