Thermal comfort is related to human health, work productivity, and building sustainability. Thus, it is inevitable that a thermally comfortable indoor environment will be created. A number of studies have been conducted using body-worn devices to achieve personal thermal comfort prediction. However, this method requires expensive equipment and lacks feasibility in a real work environment. This study proposes a predictive model using computer vision-based body thermal distribution, focusing on the purpose of extracting features from various parts of the body. It has been proven that the method proposed in this research can achieve high predictive performance without utilizing physiological factors. Body thermal distribution based on thermal images has great potential for predicting thermal comfort. Since the pixel values in the thermal image are time-dependent, the LSTM model was the most suitable deep learning algorithm. The accuracy was 98.78% showing high predictive performance when thermal image and microclimate data were used at the same time. The proposed method is valuable due to its high performance with efficient economical value and feasibility.