Motivated by the high utility and growing availability of Floating Car Data (FCD) streams for traffic congestion modeling and subsequent traffic congestion-related intelligent traffic management tasks, this paper proposes a grid-based, time-inhomogeneous model and method for the detection of congestion from large FCD streams. Furthermore, the paper proposes a simple but effective, high-level implementation of the method using off-the-shelf relational database technology that can readily be ported to Big Data processing frameworks. Empirical evaluations on millions of real-world taxi trajectories show that 1) the spatio-temporal distribution and clustering of the detected congestions are reasonable and 2) the method and its prototype implementation scale linearly with the input size and the geographical level of detail / spatio-temporal resolution of the model.