Speed Estimation and Abnormality Detection From Surveillance Cameras

Panagiotis Giannakeris*, Vagia Kaltsa, Konstantinos Avgerinakis, Alexia Briassouli, Stefanos Vrochidis, Ioannis Kompatsiaris

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingAcademicpeer-review

177 Downloads (Pure)


Motivated by the increasing industry trends towards autonomous driving, vehicles, and transportation we focus on developing a traffic analysis framework for the automatic exploitation of a large pool of available data relative to traffic applications. We propose a cooperative detection and tracking algorithm for the retrieval of vehicle trajectories in video surveillance footage based on deep CNN features that is ultimately used for two separate traffic analysis modalities: (a) vehicle speed estimation based on a state of the art fully automatic camera calibration algorithm and (b) the detection of possibly abnormal events in the scene using robust optical flow descriptors of the detected vehicles and Fisher vector representations of spatiotemporal visual volumes. Finally we measure the performance of our proposed methods in the NVIDIA AI CITY challenge evaluation dataset.
Original languageEnglish
Title of host publicationCVPR AIC 2018 - Computer Vision and Pattern Recognition Workshop NVIDIA AI City Challenge
Number of pages7
ISBN (Print)9781538661000
Publication statusPublished - 18 Jun 2018
EventIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) - UT
Duration: 18 Jun 201822 Jun 2018


ConferenceIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)




Dive into the research topics of 'Speed Estimation and Abnormality Detection From Surveillance Cameras'. Together they form a unique fingerprint.

Cite this