The canopy temperature is essential for specifying crop water stress and plays a vital role in an irrigation schedule. The canopy temperature captured from an inexpensive camera is noisy and fluctuates since the environment in practical is uncontrollable such as light reflection, wind, and rain conditions. Therefore, high accuracy canopy temperature measurement is necessary. This paper aims to estimate the canopy temperature from an IoT thermal camera device installed at a durian tree. The proposed method takes the noisy canopy temperature as the input. It uses a Kalman filter whose prediction part is substituted by a moving-average model to estimate a more-accurate canopy temperature. The experimental results showed that the proposed method could reduce the average mean-absolute error by approximately 1% and 15% compared to the reliable ground truth and the ground-truth model, respectively. Even though the improvement, when compared to FLIR, is not significant in error reduction, the proposed concept seems practical and possible to investigate further for performance improvement.