Video based people counting system

Tracking people number in a building, in a floor or in a room holds critical importance in commercial applications and in security services. In work places like shopping malls, counting the customer number is needed to estimate labor requirements and to calculate costs correctly. There are many systems, working with different types of sensors (infrared, laser, rf etc.), used for people counting. Those systems are not able to perform counting task with required accuracy and reliability because of some specific problems. Recently sensor based people counting systems are replaced by video based systems in many applications, owing to improvements in computer vision technologies. Video based systems do not block people flow and can perform counting task in different environments with high accuracy. These abilities play an important role in choice of video based systems for people counting applications. In this thesis, a video based bi-directional people counting system is developed. Designed system is able to count people with high accuracy and reliability. Computer software developed for the application uses video captured by an overhead mounted camera as input and counts people going in and out of an observed area. Even when people passing very close or touching to each other system is able to perform counting task with a very high success rate due to use of ?K-Means Based Segmentation Algorithm?. People may pass through the observed area with large suitcases, shopping cards or strollers. These objects could cause wrong counting if they are not detected. Developed system able to detect these objects and prevent wrong counting by using a human/object check algorithm. Developed system is tested with real world videos and it is observed that system is able to perform counting task even when 7 people passing through the observed area at the same time.