TY - JOUR
T1 - A blind-zone detection method using a rear-mounted fisheye camera with combination of vehicle detection methods
AU - Dooley, Damien
AU - McGinley, Brian
AU - Hughes, Ciarán
AU - Kilmartin, Liam
AU - Jones, Edward
AU - Glavin, Martin
N1 - Publisher Copyright:
© 2015 IEEE.
PY - 2016/1
Y1 - 2016/1
N2 - This paper proposes a novel approach for detecting and tracking vehicles to the rear and in the blind zone of a vehicle, using a single rear-mounted fisheye camera and multiple detection algorithms. A maneuver that is a significant cause of accidents involves a target vehicle approaching the host vehicle fromthe rear and overtaking into the adjacent lane. As the overtaking vehicle moves toward the edge of the image and into the blind zone, the view of the vehicle gradually changes from a front view to a side view. Furthermore, the effects of fisheye distortion are at their most pronounced toward the extremities of the image, rendering detection of a target vehicle entering the blind zone even more difficult. The proposed system employs an AdaBoost classifier at distances of 10-40 m between the host and target vehicles. For detection at short distances where the view of a target vehicle has changed to a side view and the AdaBoost classifier is less effective, identification of vehicle wheels is proposed. Two methods of wheel detection are employed: at distances between 5 and 15 m, a novel algorithm entitled wheel arch contour detection (WACD) is presented, and for distances less than 5 m, Hough circle detection provides reliable wheel detection. A testing framework is also presented, which categorizes detection performance as a function of distance between host and target vehicles. Experimental results indicate that the proposed method results in a detection rate of greater than 93% in the critical range (blind zone) of the host.
AB - This paper proposes a novel approach for detecting and tracking vehicles to the rear and in the blind zone of a vehicle, using a single rear-mounted fisheye camera and multiple detection algorithms. A maneuver that is a significant cause of accidents involves a target vehicle approaching the host vehicle fromthe rear and overtaking into the adjacent lane. As the overtaking vehicle moves toward the edge of the image and into the blind zone, the view of the vehicle gradually changes from a front view to a side view. Furthermore, the effects of fisheye distortion are at their most pronounced toward the extremities of the image, rendering detection of a target vehicle entering the blind zone even more difficult. The proposed system employs an AdaBoost classifier at distances of 10-40 m between the host and target vehicles. For detection at short distances where the view of a target vehicle has changed to a side view and the AdaBoost classifier is less effective, identification of vehicle wheels is proposed. Two methods of wheel detection are employed: at distances between 5 and 15 m, a novel algorithm entitled wheel arch contour detection (WACD) is presented, and for distances less than 5 m, Hough circle detection provides reliable wheel detection. A testing framework is also presented, which categorizes detection performance as a function of distance between host and target vehicles. Experimental results indicate that the proposed method results in a detection rate of greater than 93% in the critical range (blind zone) of the host.
KW - Automotive machine vision
KW - Contour detection
KW - Harris corner detection
KW - Hough circles
KW - Kalman filter
KW - Optical flow
KW - Test framework
KW - Vehicle detection
UR - http://www.scopus.com/inward/record.url?scp=84960813694&partnerID=8YFLogxK
U2 - 10.1109/TITS.2015.2467357
DO - 10.1109/TITS.2015.2467357
M3 - Article
AN - SCOPUS:84960813694
SN - 1524-9050
VL - 17
SP - 264
EP - 278
JO - IEEE Transactions on Intelligent Transportation Systems
JF - IEEE Transactions on Intelligent Transportation Systems
IS - 1
M1 - 7230256
ER -