The most significant risks that could jeopardize the success of our project is if we do not gather enough validation data/tests. We are in the process of validating our backend system and prediction, and have done testing on three different days last week. We will conduct more testing to validate our project and attempt to reduce any more errors we find.
We also mounted our camera system on top of wood poles so they could be deployed/taken away easily. This will also help us when we need to move our system to wiegand gym for the final demo. Currently our main camera system contains of two cameras, each tracking two doors in Hamershlag 1300 wing. We also setup two validation cameras that look at the same set of doors, which gives us a way to automatically validate our data rather than watching the video feed and manually validating. One of the challenges we previously met was unable to achieve a high enough fps rate for our backend processing system, leading to dropped frames and inaccurate results. This was resolved by deploying our backend code to google colab, as we have access to GPU, and could run multiple instances. We found a way to stream video to google colab through setting up web tunnels on raspberry pi as mentioned in the previous report. We did not incur any additional costs as we are using the free tier for google colab, as it is sufficient for our current setup. We are storing count data, etc. to database files stored on google drive(which could be mounted to google colab runtimes), enabling us to store the data permanently, and we could also access that data for our web application as well as prediction.
Next week, we will work on integrating the web app as well as doing more validation testing, then move on to preparing for the final demo.