Computer Vision Model for Real-Time Autonomous Wrong Way Driving and Accident Detection on Highway Roads
KEZEBOU, Landry.
2019
-
According to the
Association for Safe International Road Travel (ASIRT) [1], nearly 1.25 million people
die in road crashes each year; that is on average 3,287 death every single day, not to
mention an additional 20-50 million are injured or disabled every year. The same source
reveals that road crashes rank as the 9th leading cause of death. The National Highway
Traffic Safety Administration ... read more(NHTSA) also reported 37,461 deaths from road accidents in
2016 in the U.S alone. Existing methods for detecting wrong way drivers or accidents
rely solely on sensor networks which are susceptible to weather interference, expensive
to install, and not the most efficient in this technological age. Furthermore, none of
the existing methods have the capability of detecting both wrong way drivers as well as
accidents. We propose a computer-vision based model for detecting wrong way drivers as
well as accidents in real time, using only video feeds from already installed
surveillance cameras. The full system would then report incidents to relevant
authorities. Our model eliminates the labor-intensive task of manually analyzing
surveillance footage and presents extra features such as traffic counting. Video feeds
from surveillance cameras are fed into the system. Our flow detection algorithm maps the
road based on traffic movement, and automatically establishes direction of traffic flow;
but the direction can also be set manually. Next, a deep convolutional neural network
performs real time object detection on each frame. The centroid of each object is
tracked over subsequent frames and its direction of flow is computed. All vehicles going
opposite to the established direction of flow are flagged as wrong way and an alert or
alarm is sent to relevant authorities who can then take necessary actions. An accident
detection model is also implemented to detect accidents in real time from the same
surveillance video feed.
Thesis (M.S.)--Tufts University, 2019.
Submitted to the Dept. of Electrical Engineering.
Advisor: Karen Panetta.
Committee: Karen Panetta, James Intriligator, and Firas Said.
Keyword: Electrical engineering.read less - ID:
- jd4739535
- To Cite:
- TARC Citation Guide EndNote
