Explore a comprehensive talk on battery-optimized people counting using Far Infrared (FIR) and Artificial Intelligence (AI). Delve into the product genesis, power efficiency considerations, challenges, and limitations faced in developing this technology. Learn about AI implementation, including picture acquisition, attributing, training, and testing processes. Discover the intricacies of quality assurance in this innovative approach to people counting. Gain insights into topics such as thermal camera usage, footfall processing pipelines, mesh connections, neural network quantization, and performance comparisons. Understand the importance of calibration, network accuracy, firmware updates, and privacy considerations in implementing this technology. Examine the use of passive infrared sensors, patent applications, and uncertainty evaluation in the context of FIR and AI-based people counting systems.
Overview
Syllabus
Introduction
Sponsors
tinyML events
tinyML community
tinyML YouTube channel
next tinyML talk
moderator
Introducing Lucas David
About tinyML
About the PortalBeam
Tech spec
Challenges
First challenge
Second challenge
Data collection
Bad pixels
Noise
Temperature
Thermal camera
People counting
footfall
processing pipeline
Mesh connection
First of all we did
Did you find that it was per camera
Did you stick to 8080
Did you use a video camera
Quantization of the network
Neural network
Quantization
Performance comparison
Coordinates
Mixed precision
No other questions
No performance comparison
Why not daily
Calibration
Success
Network accuracy
Firmware updates
Passive infrared sensor
Privacy
Hiding the camera
Patent applications
Uncertainty evaluation
Training
Conclusion
Taught by
tinyML
