Lack of attention while navigating in unstructured traffic causes mishaps as road obstacles such as vehicles, jaywalkers do not follow road rules and their pattern of movement are highly unpredictable. This research work aims at developing a system that can be deployable on present vehicle for early detection and warning of road anomalies, so that evasive actions can be taken for safe navigation. We have developed near range surface obstacle sensing system using vision sensor, which can alert the driver, if any obstacle is entering the drive region or in the drive region within 15 meters. To improve real time performance of deep learning model, proposed methodology first predicts the drive region on road using monocular camera based on vanishing point (VP). YOLO (You Look At Once) darknet v3 detector is integrated with tracking algorithm SORT (Deep simple online and real time tracking) and multilevel perceptron neural network does early predication to alert the driver based on spatial information of detected obstacle. A visual alert will display the type of obstacle entering or within the drive region. The experiment result depict YOLO-SORT-MLP can ensure early detection and tracking of obstacle at real time at different time of day and weather conditions at 0.91 F1 score. The proposed methodology is been deployed on the test vehicle and tested on unstructured traffic environments.
Vision Based Driver Assistance for Near Range Obstacle Sensing under Unstructured Traffic Environment
01.12.2019
3368617 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Vision Based Driver Assistance in Urban Traffic
British Library Conference Proceedings | 2000
|Employing range imagery for vision-based driver assistance
Tema Archiv | 1998
|