Indoor occupancy estimation has become an important area of research in the recent past. This work investigates the feasibility of an Internet of Things (IoT) based university classroom occupancy estimation system. As IoT devices generate voluminous data at high rates, the centralized cloud computing approach is found to generate high latencies. The client server based cloud architecture has been compared with the decentralized edge computing architecture for building the occupancy estimation system. The performance of these architectures has been compared using two performance metrics: latency and throughput. The occupancy estimation models using carbon dioxide and relative humidity as inputs, have been developed using multiple linear regression and quantile regression. Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) have been used to compare the performance of our estimation models.