We build a real-time multi-people tracker, which is based on the Kalman Filter. The input to the software is a Probabilistic Occupancy Map of the observed area. The main goal of the project is to incorporate this tracker to the real-time detection software available on the CVLab demo room. A standalone version is also built. The algorithm exploits appearance cues to prevent identity switches. Instead of computing the appearance difference in a frame-by-frame manner, an appearance model is initially built when an individual enters the scene and is afterwards matched against the detected people. The frame-by-frame spatial tracking of the Kalman Filter makes the algorithm computationally efficient and the appearance model matching increases the robustness. The experiments performed in the demo room show that the method is satisfactory. We also validate our algorithm on a few datasets and the results prove that the method can be used in many scenarios. In certain datasets it even outperforms the state-of-the-art method while it’s one to two orders of magnitude faster.