Automated Approach for Computer Vision-Based Vehicle Movement Classification at Traffic Intersections

Movement-specific vehicle classification and counting at traffic intersections is a crucial component of various management activities. In this context, with recent advancements in computer-vision-based techniques, cameras have emerged as reliable data source for extracting vehicular trajectories from scenes. However, classifying these by movement type quite challenging, characteristics motion obtained way vary depending on camera calibrations. Although some existing methods addressed such tasks decent accuracies, the performance significantly relied manual specification several regions interest. study, we proposed an automated method movement-specific (such right-turn, left-turn through movements) vision-based trajectories. Our framework identifies different patterns observed scene using unsupervised hierarchical clustering technique. Thereafter, similarity-based assignment strategy adopted to assign incoming identified groups. A new similarity measure was designed overcome inherent shortcomings Experimental results demonstrated effectiveness approach its ability adapt scenarios without any intervention.