<![CDATA[Non-contact infrared (IR) temperature sensors on Smart Wheelchairs are important for thermal health monitoring, but they are susceptible to a significant decrease in accuracy (bias) due to non-contact factors and dynamic operational environment variations, such as outdoor exposure. This research aims to improve the accuracy of the MLX90614 IR Sensor on the SMATSI Smart Wheelchair using the Linear Regression (LR) algorithm as a computationally light calibration solution. Data collection was carried out in four crucial scenarios, including Controlled Indoor and Exposed Outdoor conditions. The initial analysis results showed that the Raw IR sensor's Absolute Error reached 0.81℃ in the outdoor scenario, which is much higher than the contact sensor's (0.24℃–0.27℃), validating the presence of extreme environmental bias. The developed Linear Regression model achieved a very high fit, evidenced by a Coefficient of Determination (R²) of 0.995, a Mean Absolute Error (MAE) of 0.056℃, and a Root Mean Square Error (RMSE) of 0.11℃. The calibration implementation successfully reduced the Absolute Error substantially in the Exposed Outdoor scenario, dropping from 0.81℃ to just 0.49℃, proving the efficacy of LR in neutralizing dynamic environmental bias. This calibration model is specific and optimal for non-contact sensors, but is not suitable for contact sensors whose baseline accuracy is already high.]]>
Copyrights © 2026
