You could do something like:
- take pictures every few seconds.
- reduce number of pixels
- convert to greyscale with standardised mean and variance
- get difference of consecutive pictures
- feed into numpy array
meanwhile record (from the pictures probably) the number of people coming into and out of the shop.
Then simply feed your data through a version of google's AI system
http://www.nature.com/nature/journal/v5 ... 14236.html or try just using some simpler correlation function. Probably not terribly accurate but might be fun.
Alternatively you could try running a face recognition algorithm on each picture and use that data instead
https://realpython.com/blog/python/face ... th-python/