Summary

We saw that appearance-based approaches start from what can be actually seen in the picture, they are not limited to the specific object classes and that they can deal with very complex objects like faces or a cars. Model-based approaches use some additional information such as shapes which were pre-defined for the recognition task.

In the above articles various methods were presented. We saw that almost every method proposed above contains the following procedure: locating the object in an image, then using feature extraction on this so called Region of Interest (RoI), and finding the approprite class from the extracted features.

The database contained at least tens of images and at most thousands of images. Also detection from videoclip was used. In [27] was presented how to use the boosting methods to detect a pedestrian to create huge database used for other related work.

We can see from the articles briefly described above that we should expect our classification rate into car makes between 80-97%.