Todays street traffic is optimized for manual driving with humans, resulting in the fact that most information transfer in traffic highly relies on visual recognition. Therefore, it only makes sense to utilize cameras for automated driving. While cameras among the other sensors provide the most information in their measurement, it is very challenging to extract this information. For an automated tram prototype a precise interpretation of the camera data is needed. One part of this problem is to detect obstacles. Within this work requirements should be identified, multiple existing algorithms for visual object detection compared. Based on this analysis one algorithm should be implemented, tested and evaluated.