Maschinelles Sehen

Beim Maschinellen Sehen geht es um die Interpretation von Kamerabildern, d.h. um die semantische Deutung des Bildinhaltes. Hierzu werden aufgrund der überlegenen Leistungsfähigkeit zurzeit fast ausschließlich Tiefe neuronale Netze, trainiert durch maschinelle Lernverfahren eingesetzt.

One way of interpreting image data is object recognition and object classification. For automated driving, for example, we distinguish between cars, trucks/buses, pedestrians, cyclists, pedestrians, traffic signs, traffic lights and many more. The detected objects are marked in the video by different colored boxes.

Another approach is the classification of each pixel, i.e. deciding to which object class the image pixel belongs. This is marked in the video by different coloring of the image. The areas of the same color mark an object class and form an object. If semantic segmentation and object formation are performed simultaneously, this is called panoptic segmentation.

Finally, depth or distance values of pixels can be estimated from mono camera images. This is shown in the third part of the video