||We present a novel stochastic recognition model based on Kernel Density Estimation (KDE) that uses a minimal set of features derived from ultrasound ranging sensors (USR) to detect presence at the desk area. In our approach, USR sensors are mounted at desk screens to provide proximity estimations of objects and users in front of them. Based on continuous proximity estimations of two screen-attached USRs, features were extracted that describe distance and motion of the user and objects. Our approach provides instantaneous presence estimation results, which is essential for energy saving, e.g. when controlling computer screens. In our evaluation with 16 users during 8 working days, we achieved a normalized recognition accuracy of more than 90%. Furthermore, we compare the KDE-based approach to established algorithms, including Nearest Centroid (NC) and Support Vector Machine (SVM). Results indicate that our KDE-based approach outperforms other algorithms using a combination of scripted data for training, and real-world recordings for testing. Finally, we show a timing study indicating that our approach is feasible to be implemented in a real-world setting.