According to relation of indoor occupancy and carbon dioxide concentration dynamic variation, design the indoor occupancy estimation algorithm based on parameter model training and Kalman filtering. The algorithm includes the model training phase and occupancy estimation phase. The model training phase uses historical measurement data to identify and describe the parameter model indoor carbon dioxide concentration variation, and the estimation phase uses the identified parameter model to design the Kalman filter to estimate indoor occupancy in real time. The test results show that the algorithm can uses historical measure data to indentify the indoor dioxide concentration variation model correctly, and estimate real time indoor occupancy when door and window are closed.