Human Detection in Thermal Images Using Transfer Learning

Recently, ChatGPT has become a hot topic, and just as AlphaGo contributed to the spread of AI in 2016, ChatGPT is now considered to have a significant cultural significance due to its direct impact on our daily lives. Furthermore, GPT's transformer also utilizes transfer learning techniques with the pre-trained GPT-3 model. In this paper, we aim to apply transfer learning to image processing, specifically creating a model for detecting humans using thermal imaging data with transfer learning applied to a pre-trained model with visible light data. In this study, the yolov5 model was used as the training model, and pretraining was performed using the coco dataset. The thermal imaging dataset for training and testing was obtained using open data from roboflow and data captured with a Lepton3.5 thermal camera. Through an experiment of detecting humans in thermal images using transfer learning, it was observed that the training speed was accelerated and the performance was improved.