Tackling the Internet of Things with Roboflow: Object Detection Apps on Android

Joo chan Kim, a PhD student at Luleå University of Technology in Sweden, is developing an Android application using object detection to identify specific IoT sensors, aiming to simplify user interaction by displaying sensor data directly when a user looks at the device. The project, intended for the Societal development through Secure IoT and Open Data (SSiO) platform, involves creating a custom model trained with 328 images of four IoT sensors from various indoor backgrounds. Data preparation included labeling and augmenting images using Roboflow to enhance diversity, ultimately achieving 557 images for model training. The team utilized SSD MobileNet V2 models for their speed and accuracy, running training on Google Colab with guidance from Roboflow's model library. Deployment on Android devices highlighted challenges such as performance in backlit conditions and on white backgrounds, prompting plans for further data augmentation and model adjustments to improve accuracy in these scenarios. The goal is to integrate this application into the SSiO platform, enabling better data interaction and service development through secure IoT and open data infrastructure.