Well, what you describe is extremely difficult to do, and people have been trying to do with for decades now without success. It could work if you place yourse

Well, what you describe is extremely difficult to do, and people have been trying to do with for decades now without success. It could work if you place yourself into similar conditions over all the videos. By having a good understanding (precise model) of what you want to recognize, you could make some pattern recognition / object detection to be able to find it in other videos. But the huge problem in this field is that the conditions keep changing. Light, luminosity, color, shape, distance. All of the makes pretty much every method of recognition of "it" completely useless. There is one thing you can currently do though. It is target tracking. Without knowing the target before, and with minimal user interaction to follow it, it is currently possible to track it quite accurately. Hope this helps.