Last week we were able to test with the live video feed, and we were able to successfully connect raspberry pi video footage to our backend CV module through CMU wifi. However, during testing with the footage, we found that sometimes there are issues with detecting people at the end of the hallway(far away) and really close(only parts of the body visible in the main entrance). There are also issues with cases of people moving fast in and out of entrances, mostly with moving out of entrances. This is because as of now, we are counting people as moving out when they get picked up as a detection and their bounding box is touching the line defining the door, thus if they walk fast, the detection algorithm only picks them up after they have moved out of the entrance, and their bounding box does not intersect with the entrance of the door, resulting in missed count. We are planning on doing more testing tomorrow, and will tweek the counting algorithm (maybe counting someone as moving out if they appear in the vicinity of the door, etc.) to see if we can handle those cases. I have also created a database file in sqlite to store count information for rooms so we can use them for prediction module, we will label the count data with additional information such as time of the day, etc, and periodically store them to the database(per minute). I think we are on schedule to have our whole system integrated by interim demo.