Increasing the reliability of sensor data, especially in collision avoidance applications, is of great importance and involves the development of different sensor fusion methods. To reduce the limitations and disadvantages of common fusion methods and their challenges with respect to highly automated driving, this paper proposes a statistical model of sensor data distribution and a new algorithm for multi-sensor data fusion according to specific detected driving situations. To this end, a scene catalogue consisting of four different traffic scenarios is modeled in a specific ADAS simulator. The analysis showed that, considering a specific road situation, the combination of sensor data as well as its training resulted in an increase in detection performance and position accuracy.