Fare-evasion detection at ticket gates using posture analysis

Ticket acquisition (either paper or digital form) and control is an integral part of Transportation Systems. Intelligent Transportation Systems (ITS) are no exception to the rule. However, the number of people using trains and subways without paying the associated fee is in augmentation each year. Fare-evasion in train station increases the feeling of insecurity for regular passengers. It also represents a loss of millions of euros every year. Hence there is a need to automatically detect fare-evasion and optimize tickets controls. Given the variety of Automatic Ticket Control Gates (ATCG), fraud characterization requires several features such as location of human body parts and body skeletal pose, interpersonal distance and movement direction. This work proposes a framework to detect fraudsters at ATCG. First it detects and segments the ATCG using Mask-RCNN. This step allows to select the area of interest, to crop it and focus only on the control gates. On this new image, the algorithm studies human postures based on skeletal extraction with OpenPose. Pose classification is used to determine if the posture is fraudulent or not.