Figure 2.
A multi-part Y O L O workflow diagram presents training, object detection, and bounding box evaluation processes. The upper-left section labelled Train process contains a flowchart beginning with Start, followed by Preparing image dataset, Labelling targets in images as ground truth, and Transform the dataset and divide it into training set and validation set. The process branches into Train model and Val model, which both connect to Y O L O model. The upper-right section labelled Object detection and Tracking process contains a flowchart beginning with Divide the grid open bracket S multiplied by S close bracket, followed by Predicted target bounding box and Confidence rating method. A decision diamond labelled Whether the threshold is met directs the process either to Non-maximum suppression for Yes or Give up bounding box for No. The accepted path continues through Output the target bounding box annotating the target type and confidence score, Central coordinates of bounding box, Displacement and acceleration calculation, and End. The middle section presents a square image divided into a 7 multiplied by 7 grid. Multiple predicted bounding boxes overlap near the centre around an object labelled structure. Labels identify Ground truth bounding box, Multiple predicted bounding box, Object structure, and Grid cell. The lower section presents three bounding box overlap examples labelled I O U equals 0.95, I O U equals 0.6, and I O U equals 0.33. Each example contains a predicted bounding box and a ground truth bounding box with decreasing overlap from left to right.YOLO algorithm: (a) YOLO flow chart; (b) bounding boxes for object detection; and (c) IOU