|M.Sc Student||Levy Ben|
|Subject||Traffic Anomalies Detection using Artificial Intelligence|
|Department||Department of Civil and Environmental Engineering||Supervisors||Dr. Sagi Dalyot|
|Professor Jack Haddad|
|Full Thesis text|
is becoming increasingly common for smart cities to collect very large volumes
over time, some of which is designed to detect unusual or anomalous data time-sequences to help better planning. For example, accurately detecting where an incident on a strip of lanes occurs will help in better resource allocation. Detection in near-real-time gives authorities the ability to maneuver traffic more wisely, thus handling traffic flow better and sometimes even saving lives.
Automatic incident detection algorithms attempt to infer the occurrence of an incident on the basis of field data that are typically measured using in-ground induction loop detectors. However, the induction loop detectors installation is complicated and expensive. The challenge of detecting traffic incidents are the sharp nonlinearities due to transitions between traffic free flow, traffic breakdown, traffic recovery and traffic congestion. Deep-learning methods have yielded tremendous breakthroughs in modeling nonlinearities while incorporating enormous data-sources of many kinds for detection or forecasting tasks. This research uses real-time spatio-temporal measurements of vehicle movement characteristics that are available from in-ground cross-sectional Bluetooth-probes across the South-North Ayalon Highway. The Bluetooth measurements are used for building an artificial-intelligence detection model that accurately identifies traffic incidents from congested traffic flows. Results of this research show the potential of employing an effective incident detection algorithm based on spatio-temporal data with the aid of Artificial-Intelligence - together with Bluetooth sensor data exploitation - as a low-cost replacement or addition for conventional existent technologies.