As we explained in the Design Review Report, one of the potential risks we have for this project is whether we will be able to achieve the 20 FPS benchmark without the GPU. While it may be too late to make the full switch to Jetson Nano, we are taking precautions with our assumptions. For one, one of the most important requirements we have is for the end users to see an updated estimation count every 1 minute, so the latency requirement leaves room for error if the FPS happens to be lagging occasionally. Once we are able to test the FPS of our detection system once the backend algorithms are integrated and implemented on top of the processed video feed, there is time allocated on our detailed schedule for us to focus on optimizing the FPS performance of the backend system.
After we made the change to our subsystems to improve the technological complexity of the project, in terms of the hardware components, we need to learn more about Transmission Control Protocol as a means to assemble the video pipeline and send the TCP packets to the backend for processing. In the backend, we will need to continue to do our research about YOLO and DeepSORT to ensure that our system can yield the best possible estimation accuracy without hurting system performance as a whole. In the frontend, it is necessary for us to learn about Django models and the Model View Controller (MVC) design pattern so that the user interface will be bug-free when a user inputs a time and is redirected to the prediction page.
We will definitely devote more time into learning about these new components caused by the changes we made to the project, but there will not be any extra material cost as of right now. We may order more ArduCAM cameras for validation purposes in the coming weeks, and we will make sure we order them early so that shipping will not pose a threat to our testing schedule. Given that we previously changed the testing spot from Sorrells to the 1300 wing, there has been small tweaks in our design requirements to ensure our system will fit in the new environment (i.e. the view of the camera, explained in detail in the design document). Next week, as we return from break, we will mostly focus on getting the backend algorithms ready for testing/integration.
Updated schedule as of March 4 (higher resolution version attached to Design Review Report):
