Question L3-C2

The question is about the differences between classical computer vision based on the engineering of feature detection and an approach using machine learning. On the right there is an illustration of a road tunnel with a dark smoke filling the tunnel and coming out from the entrance. The question is: "You work for a company in charge of the security of a main car tunnel. They want to develop a drone that can enter the tunnel in case of fire, and deliver high
resolution images of the situation. They want to use the vision also for navigation, looking at the ground lines. In case of fire the lines can be affected in very different and mostly unknown ways depending on the type of fire, etc. and what it’s important is to know when the line cannot be detected anymore by the system. Which approach you use for this computer vision task of identification of the line?". Two possibilities are proposed:

Answer A - Classical methods - is correct: The fact that the lines «can be affected in very different and mostly unknown ways» makes hard to train ML, and therefore classical methods should allow to get out the best of the image, allowing also to mesure the confidence of the result and providing a clear indication of when they cannot detect.
In the explanation of the students we would like to see that they understand that this solution allows to understand the mechanisms and control the quality of the output.

Answer B - Machine learning techniques - is wrong: The fact that the lines «can be affected in very different and mostly unknown ways» makes hard to train ML.
In the explanation of the students we would like to see that they understand that this solution requires data to be trained and that the variability of the situation does not allow it.