This paper describes a real-time people counting system

which combines the pedestrian detection and tracking method together for people counting. The normalized MB-LBP feature

which can be easily calculated and suitable for multi-scale detection

is adopted for detection in the foreground at the detecting stage. The points are grouped together with a probability model to track the pedestrian at the tracking stage. At the same time

these points are employed to map the detected pedestrian to the tracked pedestrian. And in order to verify the performance of the system

three experiments are designed for testing. The first two videos are captured at different locations with different backgrounds; the third video

which has been used for testing in many other papers

is used for comparison. The experiments results demonstrate that this system performs the function of counting well at different background situations.