madelonhulsebos / computervision Goto Github PK

• Check common evaluation metrics
• Detection component
• Segmentation component
• Regonition component
• System as a whole
• Finalize report on this aspect

For this we would like to use the Viola–Jones object detection framework, which uses AdaBoost with many different decision stump classifiers based on HAAR features.

• Decide on the project
• Write project proposal

• Een redelijke basale approach van iemand anders: http://stackoverflow.com/questions/36407608/recognize-the-characters-of-license-plate
• Paper over "Segmenting and Recognizing
  License Plate Characters": http://www.ucsp.edu.pe/sibgrapi2013/eproceedings/wuw/115776_1.pdf
• Voorbeeld van hoe iemand anders dit probleem heeft aangepakt: https://www.reddit.com/r/MachineLearning/comments/4i4j66/how_i_wrote_an_automatic_license_plate/
• Blob detection: http://scikit-image.org/docs/dev/auto_examples/features_detection/plot_blob.html (misschien niet heel nuttig omdat blobs volgens mij over regions gaan en niet zozeer variabele vormen)
• Component recognition: https://en.wikipedia.org/wiki/Connected-component_labeling#One_component_at_a_time

Write a section how we used HOG and SVM (similar to Dalal-Triggs) to detect presence of license plates.

See notes by @madelonhulsebos

At http://www.acme.com/licensemaker/ we can generate synthetical license plates. I have already written a script that downloads randomly generated license plates for various states from this site, but we may still need to do some additional work to make them usable as training samples. For example, we may need to add background noise before we can train a classifier which can identify whether or not a license plate is present in a sliding window extract.

Need to remove redundant models, redundant code/comments in the Main.py file and also remove some images. Make sure the HOG detector is being trained on the right dataset (with_background), remove incorrect references to flies etc