Fig. 5. HOG feature description in container handle detection. (a) Original image; (b) Handles description by HOG; (c) SVM classifier training. Fig. 6. You only look once (YOLO) detection process. (a) S x S grid on input; (b) Class probability map; (c) Final detections. With the development of computing resources, the use of convolutional neural networks (CNNs) to detect image features has become possible. CNNs achieve superior performance compared with traditional classifier algorithms [10], and they do not require description features from the image as observed in the HOG process. CNNs can directly extract features from the original image. We consider you only look once (YOLO) [11] as an example; its detection process is illustrated in Fig. 6. YOLO divides the original picture into S × S grids and then uses convolution calculations to extract features and detect the targets. This operation only performs one convolution calculation on the entire image; therefore, its detection speed is significantly faster than that of a detection-purpose classifier. Owing to these advantages, the CNNbased image recognition algorithms have been extensively used in recent years. Fig. 7. Basic principle of VBM. container handle. On one side of the plane H are the positive samples (container handle), and on the other side are the negative samples (these samples do not contain a container handle). The trained classifier can be used to detect the target feature area in other scenes. 68 Target Measurement Algorithm The target detection algorithm only obtains the coordinates of the target from the image. To obtain the actual position data, a conversion process is necessary. The basic method is to convert coordinate data into real geometric data. Fig. 7 shows the principle of a simple measurement system that has been called triangulation. In this system, the light reflected by the target on point G will enter the camera through the focal F; thereafter, it will make a virtual image at G1 on the charge coupled device (CCD) of the camera. CCD will record the light information IEEE Instrumentation & Measurement Magazine June 2021