Classification answers 'what', detection answers 'what and where, for each one'
Image classification takes a whole photo and outputs a single label, like 'cat' or 'street scene.' That's useful when you already know the image contains roughly one thing worth naming, but most real photos and video frames aren't that simple. A single street photo might contain 3 cars, 2 people, a dog, and a traffic light, and a classifier can only give you one answer for the entire image, which throws away almost everything actually happening in it.
Object detection solves a fundamentally different problem. Instead of one label for the whole image, it produces a list of detections, where each detection is a class label plus a location, typically a bounding box drawn around just that object. So the same street photo becomes a structured list: car at these coordinates, car at these other coordinates, person here, dog there. Detection isn't classification done better, it's classification done many times over, once per object, plus the added job of figuring out where each object's boundaries actually are.