I am conducting real-time surface inspection using object detection. However, I cannot avoid some false positives. Until I can strengthen the dataset, I need a temporary algorithm. The objects found as false positives repeat themselves, and I would like to use this repetition to create a similarity algorithm. For example, in face detection, the same scenario is demonstrated. As you can see, faces have been found, and they are similar, but their positions in the image are slightly shifted. I need to solve this without deep learning. Can you suggest an algorithm that could help?

Or is there any function inside yolo for solve it?