Abstract
In big cities today, there are many tall buildings which can accommodate a large number of people, and it has become an important issue to know how many people are currently in a building for safety reasons, for example. This has yielded a need for accurate counting of pedestrians who enter and leave such a building.
For accurate measurement of the number and the direction of pedestrians, a two-dimensional image pattern recognition method using an image pickup system such as a TVcamera system has been known. This method, however, has a problem in the real time processing if the quantity of image information obtained is too much, which is likely in many cases.
Basically, the number and the direction of pedestrians can be known by monitoring a pair of parallel lines across the pedestrian way. The authors have developed a method for bidirectional pedestrian traffic measurement which uses a pair of linear charge coupled devices. In this method, the CCDs monitor two parallel lines from above the pedestrian way and the number and the direction of pedestrians are measured based on the two-valued one-dimensional CCD image outputs.
This paper discusses the basic measurement conditions required for real time processing and for accurate measurement of multiple pedestrians passing in line, that is, the interval between the two monitor lines, the data sampling period, and the pedestrians recognition criteria.
The results showed that the interval between the monitor lines should be 8.5cm, that the data sampling period must not exceed 19.3ms, that the width along the monitor line to be recognized as a pedestrian should be 20.0 to 71.1cm when only one pedestrian passes and 23.7 to 71.1cm when two or more pedestrians pass side by side, and that the chest thickness should be 8.5 to 36.9cm when only one pedestrian passes and 12.3 to 36.9cm when two or more pedestrians pass one behined another. With these settings, the theoretical measurement error can be 0.3% or less.
