Open source reference platform for workspace management will enable equipment makers to quickly bring to market sensors that collect accurate occupancy data bringing differentiation in their service platforms.
Real estate overhead is reportedly a company’s second largest cost due to the impact of building shortages and rising prices per square meter. Yet, building occupation rates are not optimized, with as much as 40% of spaces left underutilized, according to a 2015 CBRE Workplace Strategy survey.
GreenWaves is supplying an evaluation board with an open source implementation of the design and the algorithm. The board runs on a single AA battery and can be used to evaluate the performance of the reference platform.
The complete solution for intelligent battery-powered building sensors
- Accurate, privacy-respecting, indoor people detection and counting
- Compatible with years of operation on batteries
- All platform elements supplied under permissive, open source licenses
- Flexible, programmable platform
- Reference board and firmware provides easy prototyping and evaluation
- The complete solution reduces integration and productization risk
Occupancy management reference platform demo
As you can see in the video GAP8 can reliably detect people in the room. The embedded neural network outputs the coordinates and size of bounding boxes around detected people. These can be used to provide a count of people or identify where a person is in the image.
Waking up every 5 minutes, acquiring a single image, running people detection and going to standby mode 24 hours a day 365 days a year on a 2400mA/h battery the system will operate 7.6 years.
The model accuracy on the current training data exceeds 93%.
GAPPoc-B, Occupancy Management Reference Platform
Available in our store
- Core elements
- GreenWaves Technologies GAP8 IOT Application Processor
- Lynred ThermEye infrared camera
- Bluetooth Low Energy radio module
- Reference board
- Reference people detection algorithm
- Reference board firmware
- Reference Android application displaying detected people
- All source code, neural network training code, labelled training images and demo board hardware design is provided under permissive, open source licenses