During the past week, my focus has been on testing different methods to send a stream of images to an online computer vision backend for processing. As we were contemplating using Google Colab to host our computer vision system, it was the first backend I experimented with. After conducting extensive research online, I was able to successfully display a video feed from a locally connected camera in a Google Colab session.
However, I soon realized that this local camera feed was inaccessible from a remote Colab session connected wirelessly. Upon further investigation, we discovered that this limitation is inherent in Google Colab’s networking capabilities. Since our camera system runs on a Raspberry Pi, it would be challenging to wirelessly communicate with a remote Google Colab session as it operates within a virtual machine. Consequently, we decided to avoid using Google Colab as our CV backend and instead redesign our system to allow the Raspberry Pi to stream video to a computer over WiFi for offline CV processing.
Even though we were unable to use Google Colab for our project, we are still on schedule with the camera module progress. Drawing from my experience in the Computer Vision (16-385) and Computer Networks (15-441) classes, I decided that using sockets to create a TCP connection for image transfer between the Raspberry Pi and the computer running CV would suffice to meet our product needs. Nonetheless, my primary focus for the upcoming week is on delivering the Design Review presentation and report. I will continue collaborating with my teammates to refine the camera module design to ensure seamless integration of all three systems in the future.