about img 1

Progress Report #4

The team's progress thus far in the capstone project process

Team Updates

Built MVP: network-connected, edge device with Raspberry Pi and Lepton, and cloud processing with filtering and Hough Transform detection algorithm.

Ranjini's Updates

Finished the PCB design with the help of Artur. Explored interaction between ESP32 and the flir lepton with little success. We can communicate with the ESP32 and flash code on to it but are not receiving frames from the Lepton yet. Got an MVP of the system working on the Pi with Arya’s help. The system now captures images and uploads it to S3 programmatically.

Ash's Updates

Got algorithm working using both Gaussian Mixture Model as well as thresholding. After discussing with Ioannis (CV professor), the best option seems to be using GMM and then using thresholding after. This week will be looking to implement that as well as dynamic numbers for the parameters in the Hough transform, which is the circle detection algorithm that determines the number of people in the room.

Arya's Updates

Worked with Ash to determine necessary dependencies and toolchains for cloud-side processing and computer vision. Spun-up EC2 instance to build dependencies, compile OpenCV, and create package for Lambda execution environment. Automated packaging process to realize returns on time-savings in future development. Worked with Ranjini to interface ESP32 with Lepton, with no success. Programmed Raspberry Pi to capture Lepton data and push to S3, followed by triggered Lambda execution for successful processing and reporting to CloudWatch Logs.

Pain Points

  • Successfully interfacing ESP32 with Lepton (e.g. package capture, etc.)
  • Experienced ELF Header offset error in openCV.so file. Had to rebuild OpenCV in an Amazon Linux container image to package A.L. build artifacts for execution in Lambda environment during processing.
  • Minimal false detection when applying GMM to existing algorithm.

Next Steps

  • Transitioning from Raspberry Pi to ESP32 MCU.
  • Adding Amazon Linux build artifacts to Lambda execution environment for Gaussian Mixture Model support.
  • Add preprocessing steps for successful application of GMM for detection.