| Popis: |
This work is making use of trajectory analytics for vehicular traffic study through developing and testing a system to identify outlier trajectories of vehicles. Real traffic data is analyzed to provide useful insights of traffic and generate traffic alerts to identify violating behavior, with the aim to facilitate traffic safety. Road safety is very critical issue, and this problem is faced by most people on the road. We developed a computer vision system that analyzes the trajectories of vehicles on the road to identify outlier trajectories. Outlier trajectories can be many types (non-typical/irregular behavior) on the road, such as backing up on a highway or changing lanes, turning, and other abnormal behavior. The work involves detection of vehicles and then analyzing its behavior on road. Yolov3, a TensorFlow model, is used for vehicle detection. Learning models are used to distinguish between normal and abnormal behavior. Finally, a system is developed using machine learning techniques, which is further trained using sufficient dataset examples. The performance of the developed system is tested in real-life situation, and it is found to be working quite comparable to human level detection of abnormal vehicle behavior. |
