SmartCountplus - Towards Automated Counting and Modelling of Non -Motorised Traffic with a Stand-Alone Sensor Device
Norbert Brändle, Stephan Schraml · 2010
1 ABSTRACT We introduce a novel visual counting device being a ble to automatically discriminate between participa nts of non-motorised traffic (pedestrians, bicyclists). Th e sensor elements (pixels) respond to relative ligh t intensity changes, thus avoiding conventional imaging and pri vacy issues usually raised by the public when it co mes to visual surveillance. Three-dimensional depth inf ormation is computed with the stereo principle, and the set of light intensity change events is grouped tog ether with a clustering algorithm to discriminate b etween moving objects. A classification algorithm based on descriptive features then identifies indivual part icipants of non-motorised traffic. A preliminary evaluation on a dataset with 128 passages shows a classificati on rate of 92% for riding cyclists and 100% for pedestrians for 2+1 classification, and 43-96% for 4+1 classif ication distinguishing between riding cyclists, pedestrians , walking cyclists, umbrellas and other objects.