Much research has been done in the field of pedestrian recognition by night. There are high requirements concerning the detection rates and false alarm rates of such systems. Conventional design of a pedestrian recognition system makes use of local classifiers - only a small image cutout is presented to the detector. This paper shows a method to improve detection results by incorporating contextual information in the classification task and tracking. The presented approach is used to evaluate detections of any local predecessor classifier by considering relevant contextual information. Context sources together with the output of the predecessor classifier are fused with Bayes' theorem and used as sensor likelihood for the JIPDA tracker. The use of a JIPDA tracker without contextual information makes it possible to reduce the false alarm rate of the local classifier by 65%. The incorporation of contextual knowledge leads to further reduction of the false alarm rate by 20%. The incorporation of context into pedestrian recognition and tracking considerably improves the results of a local classifier. This contribution delivers a thorough examination of the influence of one context source on the tracking result and shows how fast one track can be confirmed given contextual hints of different performance.
Incorporating contextual information in pedestrian tracking
Einführung von im Kontext stehenden Informationen bei der Erkennung von Fußgängermerkmalen
2010
6 Seiten, 5 Bilder, 11 Quellen
(nicht paginiert)
Conference paper
Storage medium
English
Incorporating contextual information on pedestrian tracking
Automotive engineering | 2010
|Incorporating Contextual Information in Pedestrian Recognition
British Library Conference Proceedings | 2009
|Deep Architecture for Citywide Travel Time Estimation Incorporating Contextual Information
Taylor & Francis Verlag | 2021
|Incorporating Equity into Pedestrian Master Plans
Transportation Research Record | 2020
|