Q92 : Vehicle backtracking for anomaly detection
Thesis > Central Library of Shahrood University > Computer Engineering > MSc > 2016
Authors:
Abstarct: Image processing has been applicant in wide variety fields and challenges in traffic control. There is a large amount of data in traffic control systems, so automatic detecting and tracking vehicles and distinguishing violation in motions will be applicably. This approach achieving by a real time intelligent system that can track and detect cars. This system also must be able to interpret the movement of vehicles for distinguish violent.
In this proposal, a system suggested that process traffic videos frxame by frxame to achieving invalid motions. At first this system reduce the size of input image for increasing speed of processing and preserving real timing. Then reduce image noise by Gaussian and Median methods for enhancing quality of image, so the performance of detection will be increased. In next step cars will be detected by AdaBoost cascade classifiers. AdaBoost extract features and train a neural network, it has acceptable speed. After detection, system find important point of cars like centers and track that points in consecutive frxames so it can achieve motion path of cars. System save this path in a matrix and process it. Next step is interpreting car motions. The legal path and motions are obvious by other cars motion and traffic rules in a scene so system can find illegal motions and violations by special simple or complicated calculations on motion matrix. The procedure accomplished on sample traffic videos. Results show that calculation precision for distinguish stopping is 68.18%, deviation to the left is 86.66%, circumvent the ban is 92.85% and for move the opposite direction is 93.61%.
Keywords:
#traffic videos #violation distinguish #tracking #preprocessing #feature extraction #AdaBoost cascade classifier #Kalman filter #traffic rules Link
Keeping place: Central Library of Shahrood University
Visitor:
Keeping place: Central Library of Shahrood University
Visitor: