Goal: experiment with different image detection strategy and see which works the best on the armor boards. You will need to work with people who work on dataset generation and benchmark.
Proposed detection strategies:
RCNN-like structure: generate proposed regions with traditional algorithms, and feed them into binary classifiers (for now, we have a LeNet. But SVM should suffice).
Pure DL approach: use DL object detection/instance(or semantic) segmentation. Note that in this case you need to think of armor boards as small objects and take that into consideration.
Pure traditional approach: DSP (Digital Signal Processing), you can ask Alvin for details.
Goal: experiment with different image detection strategy and see which works the best on the armor boards. You will need to work with people who work on dataset generation and benchmark.
Proposed detection strategies: