รหัสดีโอไอ 10.14457/TU.the.2019.756
Title A robust technique for image classification and detection for intelligent transportation system in the dark environment using spatial and temporal gradient
Creator Sorn Sooksatra
Contributor Toshiaki Kondo, Advisor
Publisher Thammasat University
Publication Year 2019
Keyword Drowsiness detection ,Gradient vectors ,Displacement vectors ,Infrared LEDs ,Face detection ,Headlight recognition ,Night-time ,Traffic surveillance system ,3D structure tensor
Abstract An intelligent transportation system has a main role to manage the traffic condition and prevent the accident or incident occurred on the road. With improvement of technology, a computer vision-based technique is able to be employed in real-time application because of low cost of installation and maintenance. Since the driver and vehicle are the main causes of a traffic accident, the camera operated in ITS is usually installed along side of the road and inside the cars for monitoring vehicles and drivers, respectively. Even though a computer vision-based technique has widely used in recent years, they are facing with severe challenges (e.g. low video quality, low illumination, various shooting condition, and so on). In this thesis, we have developed video processing techniques which are robust to various lighting conditions. Then we have attempted to apply those techniques to two applications for Intelligent Transportation Systems (ITS). Therefore, the thesis comprises two main projects: the first project is related to vehicle detection especially at night and the second project is concerned with driver’s drowsiness detection, including dark lighting conditions, where they belong to Traffic Surveillance Systems (TSS) and Advanced Driver-Assistance Systems (ADAS), respectively. It should be noted that both ADAS and TSS are part of a variety of projects in ITS. The objective for the first project is to improve traffic management by counting the number of vehicles even at night. On the other hand, the objective of the second project is to reduce traffic accidents by detecting drivers’ drowsiness and alarming them to be awake and alert.In the first project, the vehicle headlight was extracted for detecting vehicles in the night time. A common problem of this task is the similarity between the headlights from their reflections on the road. This thesis proposes a novel algorithm to construct 3D motion trajectories of headlights and their reflections on the road using both spatial and temporal information. 3D structure tensors are utilized as shape features for recognizing the headlights in various traffic views.The second project presents a drowsiness detection method for drivers based on visual features in a video sequence. In the pre-processing step, a face and a pair of eyes are detected by the Haar cascade method. We then locate a dark circular object, i.e., the pupil, within the detected eye region using two vectors: displacement vectors and gradient vectors. The normalized cross-correlation between these two vectors is maximized at the center of the circular object. The cross-correlation is also used for recognizing the state of the eye, closed or open. Infrared LEDs are always turned on for illuminating the face. Experimental results show that the proposed method works well in various lighting conditions. The computation speed of the proposed method is fast enough to perform at video rates.
ดิจิตอลไฟล์ Digital File #1

บรรณานุกรม

Sorn Sooksatra และผู้แต่งคนอื่นๆ. (2019) A robust technique for image classification and detection for intelligent transportation system in the dark environment using spatial and temporal gradient. Thammasat University:ม.ป.ท.
Sorn Sooksatra และผู้แต่งคนอื่นๆ. 2019. A robust technique for image classification and detection for intelligent transportation system in the dark environment using spatial and temporal gradient. ม.ป.ท.:Thammasat University;
Sorn Sooksatra และผู้แต่งคนอื่นๆ. A robust technique for image classification and detection for intelligent transportation system in the dark environment using spatial and temporal gradient. ม.ป.ท.:Thammasat University, 2019. Print.