Abstract:

In this study, the performance of two assistant systems developed on the basis of image processing for the detection of a part of the moving objects on the routes of the unmanned ground vehicles used as a moving platform was compared. Geographical location data provided by the global location system has been used in the development of assistant systems. In the study, the background images extracted by the models according to their geographical location were studied in two different ways as a reference model and an instant model. Reference models are obtained from default background images (non-moving objects) while instant models are obtained from real-time background images. The silhouettes identified as a result of the comparison process are classified using artificial nerve networks. 5 input and 3 output artificial nerve network (YSA) architecture were applied for classification. In the final layer of the architecture, the softmax function was used. In the artificial nerve network, each output represents three different classes of objects in pedestrian, car and bicycle. In the first method, the two-dimensional image collision method is used to balance the reference and instant image frameworks. The dynamic objects that enter the image by taking the arithmetic differences of the framework equalized with the frame difference method have been obtained. In the second method, the background images are divided into blocks. The average brightness values are used to create the model. The reference and the comparison process of instant models is carried out by calculating the Oklid distances between blocks. Experimental studies have achieved correct classification rates between 94% and 71% depending on the distance.

Keywords: