Kernel Density Estimation for Foreground Detection in Dynamic Video Processing for Unmanned Aerial Vehicle Application

The paper considers aspects of Unmanned Aerial Vehicle video stream foreground detection based on computer vision and artificial intelligence. Nonparametric approach for background subtraction based on kernel model was used as the most appropriate for usage in dynamic environment. One of the key elements of the kernel density estimation is bandwidth selection that significantly effects results of image segmentation and object detection. Several approaches of bandwidth selection and their impact on object detection performance is investigated in the paper. Otsu's method was used for threshold selection for image segmentation.