Develop a computer-vision-based classifier for trash. The system should use a live webcam and be able to (1) separate trash from the background, (2) classify the trash in one of the usual categories (paper, metal, plastic, etc.), (3) recognize toxic and/or dangerous waste (e.g. batteries), and (4) provide a comprehensive log of its decisions (e.g. confidence scores) and usage history. The system should work image-based and run on a contemporary Raspberry Pi platform. It may be based on a CNN as well as on any other technology, but reach a classification performance of at least 95%. One challenge will be the provision of an adequate training and test set.
Requires good knowledge in computer vision and edge programming. If performed as a diploma thesis, must include a comprehensive live and out-of-sample evaluation. AI-based code generation is not allowed for this task (which is the default for such tasks).