SENTRY
Project Idea
Sept 4
Sept 15
Main Components Obtained
Object Detection Complete
Oct 3
Oct 10
Target Tracking complete
3D Model Complete
Oct 18
Oct 27
PCB Arrived
Serial Communication Complete
Nov 10
Nov 17
Turret Assembled
The Spark Before the Storm
At first the idea was simple; build a sentry turret that can respond on its own. We knew a rough outline of what we needed for everything to work, but little did we know what uneasy terrain lay ahead of us. We ran on pure naïve optimism, the calm before the storm.
We ordered our main parts, tested them, and began to research software tracking. We decided to use the You Only Look Once algorithm (YOLO) for target detection and software-based tracking. YOLO can be trained on any data and it is able to efficiently identify and track any targets.
A Machine's First Heartbeat
Not All Lasers Point the Same Way
We soon came to realize how challenging it is to track targets in software with limited processing power. Our software was stuttering and had a long response time, but most of all our algorithm would detect false positives. Proper target identification is critical because we cannot afford our defence system to target the wrong thing. We set out to find a strong solution with fail-safes that would only filter for deer.
Where Progress Converges
After the midpoint of the semester was when our work was finally began to bear fruit. We finished the 3D design, optimized and increased our software's accuracy, and our printed circuit board has arrived. It was time to 3D print, put everything together and begin testing the software and hardware.
Built for More, Yet Not for Now
After performing tests we realized that our program ran too slow, so we ordered a specialized computer vision AI chip called the 13 TOPS Pi Hat. It would assist the Raspberry Pi with video processing and target recognition, so we converted our trained model to integrate with the AI chip. Unfortunately after may failed attempts at integrating the chip with our trained deer model, we had to cut our losses and continue without the chip. But not all hope was lost as during the integration process we learned other ways to speed up our software.
The Sunset Outshines the Day
Finally, after many trials and errors our turret is put together. We are able to run our software at 15 frames per second for smooth target tracking, The motors are perfectly synchronized and the laser is calibrated to aim in sync with the software. Describing our project as hard would be an understatement, but we are glad we didn't know what waited ahead of us, or else who knows if we would be willing to accept what waited for us.