A Car Counting Method in Aerial Images Based on Convolutional Neural Network

44907

Numbers of automated artificial intelligence systems that process images captured from unmanned aerial vehicles (UAV) are gradually increasing. Examples of these studies have been performed in urbanization and traffic applications. Determining the number of cars in the image is very crucial for these applications. Data preparation and labeling process is very laborious depending on the method it is performed. It takes a long time to prepare the data, especially with bounding box annotation. Preparation of data with point annotation reduces the time prepared with bounding box labeling by 4 times. In this study, a novel deep learning model that can learn vehicle counting from UAV images with data prepared with point labeling is proposed. A novel loss function has been proposed for the training of the model with point annotation. Experiments on the CARPK dataset show the competitive counting and localizing performance of the proposed method compared with existing methods that were been trained with bounding box annotations.

170

Convolutional Neural Networks , Car Counting , Aerial Imaging

Erciyes University Journal Of The Institute Of Science an‎d Technology

177

Erciyes University Journal Of The Institute Of Science an‎d Technology