Learning Vehicle Motion Patterns Based On Environment Model And Vehicle Trajectories

Traffic video analysis has turned into one of the most challenging fields in machine vision and intelligent transportation systems. Vehicle counting and classification, motion analysis and vehicle interaction understanding are some of the objectives that caused installation of cameras on intersections. As a strong basis for semantic analysis of videos, we need a model that can describe the scene in terms of zones and paths where moving objects must fit in. To gain this model a new robust approach for denoising input video is proposed that shows impressive improvement in results of zone learning and raised the success rate of correct zone detection to 93%. The motion path patterns are learned from the filtered vehicle trajectories based on learned model. The success rate of this stage is also raised to 93% because of great performance of zone learning.