Fisheye cameras provide a wider field of view than narrow-angle pinhole cameras in complex urban traffic scenes and are becoming more popular in vehicles since they are cheap and easy to handle. However, fisheye images are distorted due to strong perspective projections and are consequently unwarped for practical usage. This unwarping process decreases the image quality, especially at the image boundaries. Thus, CNNs trained on pin-hole camera images do not perform well on fisheye images.