As of right now, one of the notable challenges we face is figuring out a way to concurrently run up to 5 tracking cameras (1 main, 4 validation or 2 main, 3 validation) and maintain the performance of the CV system. On Friday, we tested running 2 videos processed by 2 cameras using only 1 device, and the performance did not change significantly from previous tests. We were also able to get the program up and running using Google Collab, which is a plus in terms of performance when compared to running the CV models locally using CPU configurations. This is proof that we have carefully thought about the contingency plans if running multiple cameras creates a toll on the success of the project.
One of the changes that we have discussed is to introduce a testing and validation system that contains 2 main cameras. Given that our main goal for both estimation and prediction is to accurately track the number of people in the main lab space, after preliminary testing, we believe that the best location to position the cameras to ensure we achieve the highest possible accuracy is in front of the entrances. If this assumption can be further validated by more extensive testing, we will make the 2 cameras that are positioned in front of the two entrances of the main room as our main cameras, and the remaining cameras in the hallway will still be useful for validation and testing purposes.
There is no significant changes to the overall schedule in terms of the features that we would like to complete in the end, but the general development/testing schedule will be as follows:
Week of Apr 3: Extensive testing and validation of CV and camera integration
Week of Apr 10: More validation if needed; completed testing/validation documents and notes; start integration with front end
Week of April 17: Finish up integration with front end; begin extended-period tests of integrated system
Week of April 24: More tests
More information regarding testing can be found in our Test Log.
Aforementioned camera setup images: Top Camera, New Camera