High-resolution imaging sensors (cameras) are basically suitable to picture the existence of persons in a setting in an electronically evaluable signal. The increasing significance of CMOS sensor technology points the way to more affordable systems for mass production. Also the brightness dynamics, which used to be a major issue of the video-based object recognition of the setting in outdoor application, will be manageable via new developments in sensor technology. The ability of a driver to recognize objects in the surroundings of the car is currently not met by any technical system due to both the superiority of the human eye as an imaging sensor and the information processing performance in the visual system. Technical vision systems are successful however, where the recognition task is highly structured and characterized by the specialization on one or a few object types (compare automated lane guidance of vehicles and automated facial recognition systems).
Goal of the project is the development and trial of a person-detector, which scans stereoscopic picture pairs of a vehicle-based camera for the existence of a person-like image pattern. The development primarily has the goal to detect children in the detection zone of the camera and generally, the application of the results to tackle the problem of the video-based detection of persons is expected and targeted. The problem definition aims at the realization of a detection module, which can be combined with the utilization of motion analysis and object tracing in the overall system.
Artificial neural networks and stereoscopic picture evaluations were utilized.
In the first project phase, methods and concepts about automated pedestrian detection in the street setting were developed. As intended, the obtained results represent intermediary results, which are to be further developed in the second project phase into a demonstrator.
Findings of the study are published in detail by a final report (German only) which is available online via https://www.tib.eu/de/suchen/download/?tx_tibsearch_search%5Bdocid%5D=TIBKAT%3A327094451&tx_tibsearch_search%5Bsearchspace%5D=tn&cHash=a5e8c470f72abcca45f9ec21a339c838#download-mark