No abstract

Today’s deep learning architectures, if trained with proper dataset, can be used for object detection in marine search and rescue operations. In this paper a dataset for maritime search and rescue purposes is proposed. It contains aerial-drone videos with 40,000 hand-annotated persons and objects floating in the water, many of small size, which makes them difficult to detect. The second contribution is our proposed object detection method. It is an ensemble composed of a number of the deep convolutional neural networks, orchestrated by the fusion module with the nonlinearly optimized voting weights. The method achieves over 82% of average precision on the new aerial-drone floating objects dataset and outperforms each of the state-of-the-art deep neural networks, such as YOLOv3, -v4, Faster R-CNN, RetinaNet, and SSD300. The dataset is publicly available from the Internet.

In recent years many methods have been proposed for eye detection. In some cases however, such as driver drowsiness detection, lighting conditions are so challenging that only the thermal imaging is a robust alternative to the visible light sensors. However, thermal images suffer from poor contrast and high noise, which arise due to the physical properties of the long waves processing. In this paper we propose an efficient method for eyes detection based on thermal image processing which can be successfully used in challenging environments. Image pre-processing with novel virtual high dynamic range procedure is proposed, which greatly enhances thermal image contrast and allows for more reliable computation of sparse image descriptors. The bag-of-visual-words approach with clustering was selected for final detections. We compare our method with the YOLOv3 deep learning model. Our method attains high accuracy and fast response in real conditions without computational complexity and requirement of a big dataset associated with the deep neural networks. For quantitative analysis a series of thermal video sequences were recorded in which eye locations were manually annotated. Created dataset was made publicly available on our website.

No abstract

No abstract