Detecting and counting people inside vehicles with little to no human input has many applications today. From assisting rescue authorities to smart transportation systems; from automatic crash response to law enforcement of High Occupancy Vehicle and High Occupancy Toll lanes. We propose a framework for counting passengers in moving vehicles based on various image features from various objects and contextual information. Each feature can be computed with state-of-the-art techniques, like Fisher Vectors, before being consolidated for a final detection score. Images from publicly available surveillance road cameras were taken to create a real-world data set, before training the convolutional neural network YOLOv3. Preliminary results show good prospective for this approach with the potential for improvements in each object detected in scene, thus improving the overall results. Future work can explore image enhancement with generative adversarial networks.
Detecting Number of Passengers in a Moving Vehicle with Publicly Available Data
Lect. Notes in Networks, Syst.
Proceedings of SAI Intelligent Systems Conference ; 2021 ; Amsterdam September 02, 2021 - September 03, 2021
2021-08-03
13 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch