During this week, my main focus was on setting up the vision system for live video integration with the computer vision backend. This involved registering our Raspberry on CMU-Device, configuring it to automatically connect to the network on startup, and verifying that I could communicate with it through my laptop. After this setup was completed, the system was able to stream the video footage from its Arudcam over a TCP connection to our laptop running the computer vision system. The video frames received by the laptop are fully interactable with our program setup once decoded, being able to be either further processed for live occupancy estimation, or saved in a recording for later analysis.
My portion of the project remains on schedule. My upcoming objective is to further smoothen the Wifi streaming of our camera system on the CMU internet network for our interim demo. There were some intermittent spikes in our camera latency during our testing this week, most likely due to network instability on the CMU Wifi. While network lag may be unavoidable due to the public nature of CMU-Device, I hope to be able to at least smoothen out some of its effects, through the integration of a frame buffer. That way, the input to our person detection and tracking systems would be smoother, potentially increasing estimation accuracy.