Optimization
Based on our testing we have been updating our vehicle detection parameters as well as the video output parameters to maximize accuracy and minimize lag and crashes. We were able to mount and test our camera system, which is able to capture and identify occupied parking spaces.
Since accuracy is a priority of ours, we switched our YOLOv5 model from small (s) to large (l). This change resulted in our accuracy in the online test to increase.
We only update cells whose values have actually changed, and catch failures gracefully. Try keeps the stream alive even if one update fails.
By separating the image into two halves, we can optimize how we handle occlusion logic
Delivery
Our final delivery for the Innovation Expo will include our camera in its temporary enclosure, video loop of the the spot status detection in different spots, and a scannable version of our app for visitors to use.
Management
Our group was within budget, as the majority of the budget was spent towards the camera and the objects enclosing the camera. Our permission to get our data in the parking garage was delayed, so our days for testing the camera system were shortened. In addition, a section of the parking garage would be better if it was closed off to create masking images for our system.
In terms of groupwork, the group was able to meet every Tuesday and Thursday, as well as some Fridays. Work this semester mainly focused on fine-tuning the YOLOv5 system as well as creating the app for our project. All camera work and the code for the camera system was limited to one person. The other four people in the group worked on the server and the app as those were both not local to one device.