The transition towards a human-centered indoor climate is beneficial from occupants’ thermal comfort and from an energy reduction perspective. However, achieving this goal requires the knowledge of the thermal state of individuals at the level of body parts. Many current solutions rely on intrusive wearable technologies, which require physical access to individuals facing limitations in scalability. Personalizing the indoor environment demands increased sensing at individual levels presenting challenges in terms of data collection and ensuring privacy protection. To address this challenge, this paper introduces a novel approach to non-intrusive personalized humans thermal sensing that can acquire personal data while minimizing the amount of sensing required. The method investigates multi-modal sensing solutions based on IR and RGB images, and it includes the development of a Machine Learning-based Human Thermo-Physiology Model (ML-HTPM). With the help of computer vision, features important for thermal comfort such as activity level, clothing insulation, posture, age, and sex can be extracted from an RGB image sequence using models such as the SlowFast network, YOLOv 7, while limited skin temperatures can be extracted from an IR image using OpenPifPaf for body parts detection. The developed ML-HTPM is based on data generated from an open-source JOS3 model after applying a prediction model based on Long Short-Term Memory (LSTM). The results showed that a human thermo-physiology model using machine learning can be trained, showing an RMSE of less than 0.5 °C in most of the local skin temperatures.