GAPPoc Concept
GAPPoc stands for GAP8 Proof of Concept. A GAPPoc platform includes an applicative board with associated documentation and firmware, focussing on one specific class of applications. It is intended to be used as a credible demonstrator of a final product.
A GAPPoc board is self-contained and embeds on a single PCB everything that is needed to demonstrate the selected use case. The objective is to make evaluation not only quick and easy but also very meaningful by making sure GAPPoc is sufficiently representative of a final system in terms of features, performance, power profile, battery life, and, to a certain extent, physical dimensions.
The GAPPoc concept was more extensively introduced in an earlier blog post: https://greenwaves-technologies.com/gappoc_for-edge_ai/ . The first member of the GAPPoc family, GAPPoc A, focused on computer vision. Recently, we have added a new member, GAPPoc B—relying on thermal infrared (TIR) sensing—to perform people detection and counting.
Figure 1: Thermal IR GAPPoc (GAPPoc B) Board Overview
Thermal IR-based GAPPoc B: A Battery-Operated Room Occupancy Management Concept Board
The TIR GAPPoc (GAPPoc B) is a new addition to the GAPPoc family, intended for, but not limited only to, indoor people detection and counting. A typical use case is the management of office occupancy. Scene capture relies on a low resolution (80×80 pixels) TIR sensor rather than a standard camera, limiting privacy concerns and removing possible issues due to lighting conditions. GAP8 processes the captured IR image locally using AI (specifically, passing it through a convolutional neural network) to count people and estimate their location. Only the final result (number of people and location) is sent over a BLE (Bluetooth Low Energy) radio link, thereby further limiting privacy issues as no infrared image, even very low resolution, is stored or sent into the cloud.
This scene capture and analysis occurs either on a periodic basis, for instance, every minute or every few minutes, or upon triggering of an optional PIR (passive infrared) motion sensor provisioned on the board.
The platform runs on an ultra-low power budget and can be operated from a single 3.6 V AA-size lithium battery.
Occupancy Management Platform Deliverables
The full platform includes the following:
- The GAPPoc B board, pre-programmed with suitable firmware implementing a full demo image capture, processing through an adequate network, BLE communication, etc.
- An Android app able to connect with the GAPPoc B board over BLE to visualize the number and location of detected people.
- All source codes, neural network training code, labelled training images, provided under permissive open source license.
- Associated documentation.
The platform is ready to be installed by a customer in the premises of their choice for evaluation. Using GAP8’s SDK, the customer can also reprogram the algorithms and software running on GAP8 according to their specific use case and needs.
Figure 2: Screenshot of Android app reporting one person in the room
Typical Platform Usage
A typical usage scheme would be as follows:
- Customer acquires a GAPPoc B evaluation kit.
- Customer runs evaluation tests out of the box.
- Once satisfied with the performance, customer designs their own product.
Turning the GAPPoc platform into a product usually involves the following:
- On the hardware side: Adapting the provided reference design to fit customer’s mechanical constraints and specific needs; for instance, replacing the BLE module by a preferred alternative or optimizing some test features.
- On the software side: Reflecting hardware changes such as different radio schemes; also, for best performance in the customer’s target use cases, retraining the neural network and tuning accordingly the embedded algorithms (e.g., weight and biases), etc.
Please bear in mind that although we provide very good quality algorithms and applicative software that are fit for reliable evaluation, the platform will nevertheless typically need some adjustments to best handle corner cases and specificities of the target usage context.
Performance
In a scenario where the platform runs 24 hours a day, waking up every 5 minutes on average to take a picture and processing it to have a result before going back to sleep, the full platform could run for over 7.5 years on a single 3.6 V, 2400 mAh AA-size battery (similar to, e.g., SAFT LS14500). This includes all on-board components, from GAP8 to memories to the TIR sensor, except the radio module as it is very use case dependent (type of radio, target range, etc.). The model accuracy on the current training data exceeds 93%. With such performance, the proposed system delivers up to 20 times the power efficiency of comparable solutions.
Learning More and Ordering GAPPoc B
Through the following link to our website, you can obtain further information on GAPPoc B:
https://greenwaves-technologies.com/applications_2/occupancy-management/
It includes a product brief and two short videos that are interesting:
- An introduction by Lynred and GreenWaves Technologies of the perspectives offered by their ThermEye sensor coupled with our ultra-low power GAP8 AI processor.
- A live demo of installation and operation of the GAPPoc B platform in a small office.
Both these videos are also available on our YouTube channel: www.youtube.com/c/GreenWavesTechnologies)
GAPPoc B is available for online purchase in our store: https://greenwaves-technologies.com/product/gappoc-b-occupancy-management-reference-platform/