{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/traffic4d-single-view-reconstruction-of","title":"Traffic4D: Single View Reconstruction of Repetitious Activity Using Longitudinal Self-Supervision","arxiv_id":null,"date":"2021-07-16","proceeding":"IEEE Intelligent Vehicles Symposium 2021 7","authors":["Fangyu Li","N. Dinesh Reddy","Xudong Chen and Srinivasa G. Narasimhan"],"abstract":"Reconstructing 4D vehicular activity (3D space\r\nand time) from cameras is useful for autonomous vehicles,\r\ncommuters and local authorities to plan for smarter and safer\r\ncities. Traffic is inherently repetitious over long periods, yet\r\ncurrent deep learning-based 3D reconstruction methods have\r\nnot considered such repetitions and have difficulty generalizing\r\nto new intersection-installed cameras. We present a novel\r\napproach exploiting longitudinal (long-term) repetitious motion\r\nas self-supervision to reconstruct 3D vehicular activity from a\r\nvideo captured by a single fixed camera. Starting from offthe-shelf 2D keypoint detections, our algorithm optimizes 3D\r\nvehicle shapes and poses, and then clusters their trajectories in\r\n3D space. The 2D keypoints and trajectory clusters accumulated\r\nover long-term are later used to improve the 2D and 3D\r\nkeypoints via self-supervision without any human annotation.\r\nOur method improves reconstruction accuracy over state of\r\nthe art on scenes with a significant visual difference from the\r\nkeypoint detector’s training data, and has many applications\r\nincluding velocity estimation, anomaly detection and vehicle\r\ncounting. We demonstrate results on traffic videos captured\r\nat multiple city intersections, collected using our smartphones,\r\nYouTube, and other public datasets.","url_abs":"http://www.cs.cmu.edu/~ILIM/projects/IM/TRAFFIC4D/","url_pdf":"https://www.ri.cmu.edu/wp-content/uploads/2021/05/Traffic4D_Longitudinal_iv2021.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"traffic4d-single-view-reconstruction-of","repo_url":"https://github.com/Emrys-Lee/Traffic4D-Release","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"3d-reconstruction","task_name":"3D Reconstruction"},{"task_slug":"anomaly-detection","task_name":"Anomaly Detection"},{"task_slug":"autonomous-vehicles","task_name":"Autonomous Vehicles"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}