The office recycling project is a platform to encourage and promote recycling among office workers through gamification. This project is aimed to reduce the percentage of un-recycled garbage that is produced in the offices, especially for the big companies, thus reducing the related costs and environmental taxes.
The goal is to create a competition between departments, floors, or offices, engaging the employees with rewards for the winners like extra days off, or promoting healthy habits like free fruit on Fridays. This encourages the employees and enhances the social relations among them, working as a team to obtain a goal. Furthermore, this improves the ecological footprint of the company, one of the current tendencies in corporate social responsibility, creating a positive image for the company.
The system is based on stations, where each one has a device that measures the weight of the bin containers that, with a previous calibration, is able to measure the real garbage that is recycled. The relation between this value and the un-recycled garbage that is generated in the department/office gives certain punctuation. This information is sent to a cloud infrastructure that manages the information of the whole company (all the stations).
Also, two tools help the employees to classify the waste: a touchscreen and a camera. When unused, the display shows information related to the current ranking or recycling tips, and also allows searching for a specific object type to get information about where to put it. Furthermore, there is a camera that scans the objects and gives tips about the proper recycling bin where to put them.
The hardware of each device is based on a main ST’s STM32F746 SoC, an ESP32-D0WD coprocessor that provides WiFi connectivity, a strain gauge and its analog conformation (Texas instruments) to be able to measure the weights, and external memories to handle all the features of the system. Also, it is connected to both an LCD-TFT touchscreen and a camera. It also provides a good balance between computational performance and low power consumption
To manage the image recognition of the garbage, it uses the TensorFlow Lite for Microcontrollers to be able to implement the edge machine learning. With this, we avoid the dependency on a cloud-based solution (and the infrastructure economic costs) and also get a smaller model size and reduced computational needs.
All data is sent and stored on AWS, where an ad-hoc microservices-based cloud architecture has been developed and deployed by Cactus.
There are two front-end dashboards. The first one is embedded into the hardware device, rendering the UI that is displayed on the touchscreen, and was developed using ST’s TouchGFX. The second is a web page implemented with React that shows the current status of the whole system and allows checking the recycling information of each station as well as other interesting KPI’s.