Development of a computer vision based system to detect dynamic objects for unmanned ground vehicles

In this study, a computer vision based and Global Positioning System (GPS) assistance supported system was developed for unmanned ground vehicle (UGV) or similar moving platforms to detect dynamic objects along its route. Before the real time application of the UGV, image models were created that represents a default background in all specified coordinates on the route. Five different approach were developed to obtain the image models. These models are sorts of feature matrixes which are much smaller than complete image matrices. The model matrices were recorded in the system database and a database relation was created between the model and its coordinate. The model belongs to related coordinates and feature matrices of instant images captured when the UGV moved are matched. When unmanned ground vehicle begins real-time motion, moving objects are determined in the scene by comparing the model of reference associated in memory with snapshots models. If the feature matrices are not compatible with each other, there is a possibility of dynamic objects in the scene. The artificial neural network techniques are used for recognition and classification of objects. Developed approaches provide alternatives to compensate for the noise generated by motion instead of High computational cost such complex operation. Object detection and recognition are observed to be done with low-cost computer systems and cameras in high success rate.