Reduce false positives with pre-trained image classifier

[ian-whitestone]

• [x] Layer on PIR motion sensor
• [ ] Try using image classifier (reference below)

References:

• https://www.pyimagesearch.com/2015/11/09/pedestrian-detection-opencv/

• https://www.pyimagesearch.com/2017/09/11/object-detection-with-deep-learning-and-opencv/

• https://www.pyimagesearch.com/2018/08/13/opencv-people-counter/

• https://www.analyticsvidhya.com/blog/2018/07/top-10-pretrained-models-get-started-deep-learning-part-1-computer-vision/

• https://www.learnopencv.com/keras-tutorial-using-pre-trained-imagenet-models/

• https://stackoverflow.com/questions/34871294/full-body-detection-and-tracking-using-opencvpython-2-7

• https://medium.com/@madhawavidanapathirana/https-medium-com-madhawavidanapathirana-real-time-human-detection-in-computer-vision-part-1-2acb851f4e55

[ian-whitestone]

Ignore files from 2018-10-31 < 11:15pm ET, min area was set too low full body wasn't captured in frame.

[ian-whitestone]

talk about training data biases....only tagging based off what gets triggered initially (i.e. min area > X, consecutive # of motion frames etc...), so everything you measure (like mean PIR value), will be biased based on those settings.

thing i didnt explore: triggering alerts based just of PIR sensor and then tagging as occupied/unoccupied.

[ian-whitestone]

Backtesting should save the background avg frame, so you can tune the other settings (like delta_thresh)