Automatic feature classification for object detection based on motion analysis

In this paper an approach for automatic feature classification based on their motion in the image plane is introduced. By combining concepts of the human perception of motion with techniques belonging to the area of cluster analysis, we subsequently abstract the visual data in order to separate features, whose motion is caused by the sensor motion from features, which possibly belong to dynamic objects in the environment. The presented algorithm exclusively works on data, that can be extracted from the two dimensional image plane. Hence, no external data like the current motion of the applied camera is required. Furthermore, the algorithm works on any type of tracked feature, as long as it can be statistically represented. The results of the presented approach constitute a very good starting point for additional object detection mechanisms.