We wanted to build a low-cost fall and activity camera system that can benefit society through analysis of human poses.
What it does
IOT enabled, personalized, and noninvasive Human pose estimation through OpenPose DNN Machine learning algorithm for classification Cloud enabled hardware with Google Cloud
How we built it
We designed a low-cost and home-safe camera that connects the camera from the Odysseus to a computer and then we utilize a Particle microcontroller to enable cloud-based servo motor commands and compute trajectories for the head of Odysseus to track someone in its sight to enable falling detection of anyone in the room.
- Flask Web Server deployed on Google Cloud Platform
- Camera Client runs locally Runs OpenPose [Deep Learned Human Pose estimation model] Runs our manually trained multi-class SVM from outputted pose keypoints. This classifies the pose from [‘Standing’, ‘Sitting’, ‘Fallen’, ‘Background’] PID Controller determines angle to track human in the frame, which is posted to the Web Server. Web Server communicates with the Particle Photon (IOT device) to rotate servo using Pulse Width Modulation (PWM) signal.