A Combination of Model-Free Methods and Gaussian Mixture Models for Object Detection and Recognition in Video-Based Surveillance Systems in public Transportation Scenarios

In this paper we present a video-based real-time system for detection and recognition of moving objects in transportation related scenes. Our goal is to build an intelligent video understanding system which functions without object models that need a huge number of training samples. This system should have a very high recognition rate and comply with real-time constraints. To achieve this goal, we have used multiple Gaussian mixture models (GMMs), contour moments and contour trees. Due to the robust segmentation of moving objects while involving GMMs the system performs well even under small background movements conditions. Two test scenarios are considered. The first one is on a parking place and the second one is on a highway. Under both scenarios a recognition rate of 91% has been achieved. The level of complexity of the system is low so that it can run on a chip to produce intelligent cameras for airports, streets and highways. Fadi Al Machot, Smart System Technologies, Transporation Informatics, Alpen Adria University, Klagenfurt, Austria , e-mail: [email protected] Kyandoghere Kyamakya, Smart System Technologies, Transporation Informatics, Alpen Adria University, Klagenfurt, Austria e-mail: [email protected]