No abstract

Object detection and recognition in aerial and remote sensing images has become a hot topic in the field of computer vision in recent years. As these images are usually taken from a bird’s-eye view, the targets often have different shapes and are densely arranged. Therefore, using an oriented bounding box to mark the target is a mainstream choice. However, this general method is designed based on horizontal box annotation, while the improved method for detecting an oriented bounding box has a high computational complexity. In this paper, we propose a method called ellipse field network (EFN) to organically integrate semantic segmentation and object detection. It predicts the probability distribution of the target and obtains accurate oriented bounding boxes through a post-processing step. We tested our method on the HRSC2016 and DOTA data sets, achieving mAP values of 0.863 and 0.701, respectively. At the same time, we also tested the performance of EFN on natural images and obtained a mAP of 84.7 in the VOC2012 data set. These extensive experiments demonstrate that EFN can achieve state-of-the-art results in aerial image tests and can obtain a good score when considering natural images.

In the field of UAV information processing and aerial image processing technology, data acquisition with MUAV platform is more and more widely used in emergency rescues. The traditional way of aerial photogrammetry, which has a limited accuracy in aerial camera locations on the UAV, makes measurements of the object location on pairwise images inaccurate. The method presented in this paper fast constructs a stereo aerial image aided with high-precision GPS and INS data. It also allows users to do correct measurements based on the image as the high-frequency DGPS promoting the reliability of camera locations. Dense shooting stations make a longer measurement baseline possible. The main steps to construct a stereo aerial image are: 1) Image acquisition and overlap calculation. Make a route planning for the flight according to the flight area to get an adequate overlap of the image sequence. 2) Fast stitching of aerial images. Highprecision GPS and INS data are used to roughly pre-align images of the flight area and then images are evenly stitched to two panoramas. 3) Stereo aerial image synthesis. We adjust the parallax of the stereo image, synthesized by the two, to a scale properly fitting the human eyes to make it more dynamic. The stable low-altitude flying of the MUAV as well as the high density data acquired enable us to measure objects in the stereo image with a longer baseline. In order to verify the practicability and the measurement accuracy of this method, an experiment was done along the Kelan River in Xinjiang. We obtained high-resolution images and accurate camera locations with the MUAV platform. Afterwards, a stereo aerial image was quickly constructed and measurements were taken by improved aerial photogrammetry to get detail information of the image. In the accuracy analysis of the measurements results, it demonstrated fine efficiency and a good accuracy of the method.

This paper reviews the NTIRE 2022 challenge on efficient single image super-resolution with focus on the proposed solutions and results. The task of the challenge was to super-resolve an input image with a magnification factor of ×4 based on pairs of low and corresponding high resolution images. The aim was to design a network for single image super-resolution that achieved improvement of efficiency measured according to several metrics including runtime, parameters, FLOPs, activations, and memory consumption while at least maintaining the PSNR of 29.00dB on DIV2K validation set. IMDN is set as the baseline for efficiency measurement. The challenge had 3 tracks including the main track (runtime), sub-track one (model complexity),