Practical Implementation of Upgraded Low-Cost Sensors in Everyday Home Devices

Abstract

The crucial part of IoT-controlled devices is the collection of accurate data. However, manufacturers often use low-cost sensors to make everyday home devices affordable, which can compromise accuracy. Therefore, we introduce a novel framework designed to improve the calibration performance of low-cost sensors incorporated into these devices. Applying this framework to home appliances makes it possible to calibrate low-cost sensors with inference speeds comparable to linear models while achieving accuracies similar to those of deep learning models. Specifically, the framework offers a selection of three different model variants, each considering factors such as implementation difficulty, calibration accuracy, or inference speed. Experimental findings indicate that our framework exhibits superior performance in both general-purpose and embedded hardware, highlighting its potential applicability to everyday home devices such as IoT-controlled appliances.