A bounding box is a rectangular frame drawn around an object within an image or video frame, used to mark its location and boundaries for the purposes of object detection or image analysis. It's typically defined by coordinates, often the pixel positions of two opposite corners, that specify exactly where the box sits within the larger image.
Bounding boxes are a foundational building block in computer vision: they're used both to label training data, showing a model where a particular object appears in thousands of example images, and as the output format many object-detection models return, drawing a box around each object they identify along with a label and confidence score. Getting the box tight and accurately positioned matters directly for how well downstream tasks, like counting objects or tracking their movement across video frames, actually perform.
