In this paper, we propose a human sleep detection system based on a low-resolution thermal camera and Adaptive-Network-Based Fuzzy Inference System (ANFIS). The proposed system is implemented on a microcomputer and detects a human sleeping by using thermal camera and checking the human state for a certain period of time. If a dozing off situation is detected, the system uses a buzzer to send alert sound to prevent the situation. We prepared the training dataset for ANFIS by defining sleep and other postures. From the experimental results, we found that the proposed system can classify whether a user is in a sleeping state or not with high accuracy.