Motion tracking using fuzzy logic and consistent labeling for multiple objects in multiple cameras vision

Abstract

In this work we address the issue of tracking multiple objects in an environment covered by multiple cameras having overlapped fields of view (FOV); we use FOV lines to achieve consistent labeling. The proposed method results advantageous in video-surveillance of indoor and outdoor environments (e.g., airports, metropolitan areas, banks, etc.) or in cinematic and television field for the optimal camera shooting management, suggesting the camera that offers the best view of the objects in the scene. We propose a system infrastructure and a new approach based on a Hierarchical Multilevel Clusterized Architecture (HMC) applied to image tracking. The areas that two or more fields of view have in common are the core for our approach: thanks to FOV lines detection, a Consistent Labeling mechanism is achieved with the use of projective invariants. The user has the chance to indicate one or more Interest Areas. Object Tracking is performed with the application of two tracking algorithms in parallel, ensuring efficient solutions even in scenarios with noised images. The system is time continuous: the temporal distance between two consecutive DAQ frames is much smaller than the one between two consecutive tracking frames. The Human Behaviour Analysis (HBA) Module avails itself of a Decision Support System (DSS) to guide the user in those critical situations that can occur while observing the scene. The anomaly detection is executed with the use of optional ribbons; collisions are managed through the use of the color histograms of the ribbons representing the object. The black areas management (i.e., uncovered areas) is performed by the Fuzzy Prediction System (FPS), structured on two levels and capable to predict the position of the tracked object in the instants immediately subsequent to its entrance in a black area, by means of fuzzy inference rules. The system results particularly efficient thanks to the implementation of a Consistent Labeling mechanism, united with a FPS that contributes to more and more accurate tracking. The use of a DSS and a HBA module simplifies the usage of the system, preserving its security issues.