1. UKM Learning and Research Repository
2. Master Thesis / Tesis Master
3. Faculty of Engineering and Built Environment / Fakulti Kejuruteraan dan Alam Bina

Title:	Onboard vision based driver assistance systems for intelligent vehicles
Authors:	Mohammadsaleh Javadinobandegani (P58195)
Supervisor:	M. A. Hannan, Prof. Dr.
Keywords:	Vehicle system
Issue Date:	3-Apr-2014
Description:	Road transportation is one of the primary modes of transportation throughout the world. Safety, energy consumption, oil prices, environmental concerns and many other considerations make road transportation research important. People spend a significant amount of time in their vehicles and on the road every day, and it is important to make this time as safe, pleasant and short as possible. Onboard vehicle sensors, especially visual sensors such as camera can generate valuable data that can be used to interpret the surrounding environment for further control processes in intelligent vehicles such as lane departure warning, traffic sign detection, pedestrian detection, obstacle detection, etc. Robustness and accuracy are essential for these systems due to the change in background and illumination. The nature of outdoor environment and moving cameras make object detection very complicated under the circumstances. Therefore it is necessary to develop a novel classifier particularly for driver assistance purposes. The aim of this research was to enhance driver assistance systems to help drivers to make better decisions on the road. Also it was necessary to investigate a theoretical framework for implementing driver assistance system structure using inter-vehicle communication. Inter-vehicle communication systems enable digital data communication among vehicles. Therefore equipped vehicles have the advantage of receiving other vehicles’ sensors data ahead of time. This research presented some onboard vision based driver assistance applications that can be employed for intelligent vehicles. First, a lane boundaries detection model was developed. It was based on inverse perspective mapping, edge detection and fitting lines algorithm. The system was tested on the urban road image database in different light conditions. Second, a novel and precise classifier was designed and developed for on-road pedestrian detection. This classifier was trained by the Gentle AdaBoost algorithm employing decision trees as inner loops using Haar-like features. This learning machine used 7000 positive samples including pedestrians and 3000 negative samples including non-pedestrians, to train the system. Then a vision based pedestrian detection and localizing system was introduced. This system consisted of the developed classifier, localizing and alerting systems for the driver to avoid any collision between the vehicle and pedestrians. The detection rate of the lane boundaries detection system is 97.20%. The result shows that the system is accurate and robust with respect to the shadows and worn lane markings and also appropriate for real time procedure. The proposed pedestrian classifier achieved the detection rate of 96.50% and false alarm rate of 1.10% on the test set which is better than other existing pedestrian classifiers as it is shown. The pedestrian detection and localizing system also performs significantly better than existing human body detectors which fail in detecting pedestrians crossing the street because of different point of view and angle. The proposed system accomplished with the detection rate of 94.73% and false alarm rate of 0.03%. It shows that these systems have got a promising performance to fulfill the objectives and it can be employed as a part of larger intelligent vehicle system. Onboard driver assistance systems can enhance the awareness of drivers on the road and consequently decrease the accidents enormously. The ultimate goal of this research area is to develop autonomous vehicles.,Master/Sarjana
Pages:	97
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Engineering and Built Environment / Fakulti Kejuruteraan dan Alam Bina

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.