Letting computers analyze camera images is challenging work that requires profound knowledge about signal processing and computer vision. Over the course of many years, while developing autonomously driving robots, we have acquired extensive knowledge about these topics. Starting with image segmentation (separation of an image into multiple parts that share some common characteristics) and object detection (detecting which parts of the image might be an object of interest), up to feature extraction and classification (a means of “recognizing” locations), all these methods can be combined to extract the information that is needed from the images.

Fig. 1: Performing pattern classification techniques on image segments

For instance, when using pattern analysis for detecting a certain type of object such as human beings, any candidate pattern that has been found in an image segment is then put to test by means of one or more neural networks, leading to a classification of the object as being either a match or a mismatch for the given pattern.

Fig. 2: The neural network determined that the candidate (shown in the upper left corner) is indeed a human being

The neural networks can also be adapted to special customer needs and thus be expanded to handle different types of patterns.