A Feature Level Fusion Approach for Object Classification

A new feature level fusion approach for object classification is introduced. The system is implemented to fuse sensor data of a laser scanner and a video sensor. A new method of video feature extraction incorporates features, which are obtained from the laser scanner, to handle the problem of multiple views of cars. The laser scanner's estimates of contour information can identify the discrete sides of rectangular objects. These object sides are transformed to the video image. A perspective reconstruction compensates deformations as well as size differences in the video image. Afterwards, an object detector is applied. A new method performs a feature extraction from this detector. The classification algorithms fuse these new features with additional features, which are obtained from the laser scanner and the tracking algorithms. The complete system is applicable in real time. An evaluation with labeled real world test data is given.