The image classification capabilities in DataRobot can be used to perform that type of analysis. However, the starting point is having a dataset for training the model. I have found it difficult to find publicly available datasets that show car damage.
One of my collegues may have such a dataset that shows no damage, minor damage, and major damage. I am looking for it now and will update this post with that info. Once you have the dataset, then you can use DataRobot to train a model to recognize the severity of damage. The image classification in DataRobot easily handles a 100 categories. If you have a labeled dataset, you can then test out the performance.
How does that sound?