Use this URL to cite or link to this record in EThOS:
Title: Vision-based driver behaviour analysis
Author: Yan, Chao
ISNI:       0000 0004 6059 3661
Awarding Body: University of Liverpool
Current Institution: University of Liverpool
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Access from Institution:
With the ever-growing traffic density, the number of road accidents is anticipated to further increase. Finding solutions to reduce road accidents and to improve traffic safety has become a top-priority for many government agencies and automobile manufactures alike. It has become imperative to the development of Advance Driver Assistance Systems (ADAS) which is able to continuously monitor, not just the surrounding environment and vehicle state, but also driver behaviours. Dangerous driver behaviour including distraction and fatigue, has long been recognized as the main contributing factor in traffic accidents. This thesis mainly presents contributing research on vision based driver distraction and fatigue analysis and pedestrian gait identification, which can be summarised in four parts as follows. First, the driver distraction activities including operating the shift lever, talking on a cell phone, eating, and smoking, are explored to be recognised under the framework of human action recognition. Computer vision technologies including motion history image and the pyramid histogram of oriented gradients, are applied to extracting discriminate feature for recognition. Moreover, A hierarchal classification system which considers different sets of features at different levels, is designed to improve the performance than conventional "flat" classification. Second, to solve the effectiveness problem in poor illuminations and realistic road conditions and to improve the performance, a posture based driver distraction recognition system is extended, which applies convolutional neural network (CNN) to automatically learn and predict pre-defined driving postures. The main idea is to monitor driver arm patterns with discriminative information extracted to predict distracting driver postures. Third, supposing to analysis driver fatigue and distraction through driver's eye, mouth and ear, a commercial deep learning facial landmark locating toolbox (Face++ Research Toolkit) is evaluated in localizing the region of driver's eye, mouth and ear and is demonstrated robust performance under the effect of illumination variation and occlusion in real driving condition. Then, semantic features for recognising different statuses of eye, mouth and ear on image patches, are learned via CNNs, which requires minimal domain knowledge of the problem. Finally, works on pedestrian subject identification using convolutional neural networks( CNNs) and multi-task learning model(MTL), is presented additionally. Gait identification is strongly motivated by the demands of security that require automatically identifying person at a distance. This be particularly relevant with respect to police/detective vehicle that is tracking criminal.
Supervisor: Zhang, B. ; Coenen, F. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral