So, I have decided to give up on OpenCV and haar classifiers for the algorithm and am going to use YOLOv3 and Darknet, a lightweight real time object detection algorithm. I couldn’t manage to get the FPS I wanted with OpenCV (and after much digging, figured out that on the Jetson Nano, it is almost impossible to get good results and utilize the GPU since OpenCV is not made for it) and the accuracy with Haar classifiers was lacking. I am currently working on it in YOLOv3, and am currently training the model with datasets. For the demo on Monday, I may show the Haar classifiers algorithm version if I am not done with this part by then.
However, this week Minji, Jiamin and I went to the labs and debugged our temperature sensor! We managed to get it working and measuring temperature on the Nano (very annoying, i2cdetect is super buggy on Nano) after 4 hours. The sensor just wasn’t being detected by the Nano, and we tried debugging it with an Arduino and a Raspberry Pi. We got it in the end, and the video will be posted in our team status report.
I am a little behind regarding the facial detection algorithm since I faced many roadblocks, but I am certain that Yolov3 is the way to go. I will be catching up this week since I don’t have many assignments and will be working lots with the team.
Update: Yolov3 works!