PESMOD: Small Moving Object Detection Benchmark Dataset for Moving Cameras

This paper presents a new high resolution aerial images dataset in which moving objects are labelled manually. It aims to contribute to the evaluation of the moving object detection methods for moving cameras. The problem of recognizing moving objects from aerial images is one of the important issues in computer vision. The biggest problem in the images taken by Unmanned Aerial Vehicles(UAV) is that the background is constantly variable due to camera movement. There are various datasets to evaluate motion detection methods in the literature. The prepared dataset consists of challenging and high resolution images containing small targets compared to other datasets. The dataset consists of different images such as people walking in nature, people skiing, vehicles driving through the forest, towns and vehicles going on the highway. This study also represents a post-processing method that can be used after background modeling for moving object detection. Experimental studies show that proposed post-processing technique provides a significant increase in precision and F-score metrics. The results of different methods in the literature are compared in detailed for each sequence in the dataset. The dataset and source codes are available at: http://github.com/mribrahim/PESMOD