Fine-grained Online Activity Detection for Smart Transportation and Mobility
Mentor: Roberto Javier López Sastre
Phone: (+34) 918856720
University: Universidad de Alcalá
Partner Host Institution: N.A
Keywords: computer vision, artificial intelligence, deep learning, smart transportation, Activity detection

Fine-grained Online Activity Detection for Smart Transportation and Mobility

The main target of this research project is to deploy robust solutions for automatic activity detection in extended videos. Typically, an advanced intelligent transportation system needs to work with long videos, that normally arrive in an online fashion. These extended videos contain significant spans without any activities and intervals with potentially multiple concurrent fine grained activities. Therefore, it results fundamental: 1) to research on novel online methods for the task of activity detection, that can efficiently discriminate between background and action time slots; and 2) to implement solutions for the problem of concurrent fine-grained activity detection, which implies to build approaches that can localize multiple concurrent actions in the video stream, keeping a fine granularity for the categorization. This fine-grained capability will play a fundamental role to envision the new era of smart transportation solutions, where the artificial intelligent agents will be able to interact and interpret the environment in a fine-grained fashion, hence extracting more semantic information. As an example, the solutions to be developed, will automatically detect whether a pedestrian is pulling a bag approaching to a car, then opening the door, etc. In this way, the intelligent agent can predict that the vehicle is about to depart.

Departament: Signal Theory and Communications
Research Group: GRAM
More Information:
Relevants projects on the area: Project 1: STIMULO (IPT-2012-0808-370000)
Relevants publications on the area: 1.- Embarrassingly Simple Model for Early Action Proposal. Marcos Baptista-Ríos, Roberto J. López-Sastre, Franciso Javier Acevedo-Rodríguez, Saturnino Maldonado-Bascón. Published in the Anticipating Human Behavior Workshop, ECCV 2018.
2.- Unsupervised learning from videos using temporal coherency deep networks. C. Redondo-Cabrera and R. J. López-Sastre. Computer Vision and Image Understanding, 2019.