Autonomous navigation of drones using computer vision has achieved promising performance. Nano-sized drones based on edge computing platforms are lightweight, flexible, and cheap; thus, they are suitable for exploring narrow spaces. However, due to their extremely limited computing power and storage, vision algorithms designed for high-performance GPU platforms cannot be used for nano-drones. To address this issue, this paper presents a lightweight CNN depth estimation network deployed on nano-drones for obstacle avoidance. Inspired by knowledge distillation (KD), a Channel-Aware Distillation Transformer (CADiT) is proposed to facilitate the small network to learn knowledge from a larger network. The proposed method is validated on the KITTI dataset and tested on a Crazyflie nano-drone with an ultra-low power microprocessor GAP8. This paper also implements a communication pipe so that the collected images can be streamed to a laptop through the on-board Wi-Fi module in real-time, enabling an offline reconstruction of the environment.
The increased socio-economic significance of landslides has resulted in the application of statistical methods to assess their hazard, particularly at medium scales. These models evaluate where, when and what size landslides are expected. The method presented in this study evaluates the landslide hazard on the basis of homogenous susceptible units (HSU). HSU are derived from a landslide susceptibility map that is a combination of landslide occurrences and geo-environmental factors, using an automated segmentation procedure. To divide the landslide susceptibility map into HSU, we apply a region-growing segmentation algorithm that results in segments with statistically independent spatial probability values. Independence is tested using Moran's I and a weighted variance method. For each HSU, we obtain the landslide frequency from the multi-temporal data. Temporal and size probabilities are calculated using a Poisson model and an inverse-gamma model, respectively. The methodology is tested in a landslide-prone national highway corridor in the northern Himalayas, India. Our study demonstrates that HSU can replace the commonly used terrain mapping units for combining three probabilities for landslide hazard assessment. A quantitative estimate of landslide hazard is obtained as a joint probability of landslide size, of landslide temporal occurrence for each HSU for different time periods and for different sizes.
Unmanned aerial vehicles offer a safe and fast approach to the production of three-dimensional spatial data on the surrounding space. In this article, we present a low-cost SLAM-based drone for creating exploration maps of building interiors. The focus is on emergency response mapping in inaccessible or potentially dangerous places. For this purpose, we used a quadcopter microdrone equipped with six laser rangefinders (1D scanners) and an optical sensor for mapping and positioning. The employed SLAM is designed to map indoor spaces with planar structures through graph optimization. It performs loop-closure detection and correction to recognize previously visited places, and to correct the accumulated drift over time. The proposed methodology was validated for several indoor environments. We investigated the performance of our drone against a multilayer LiDAR-carrying macrodrone, a vision-aided navigation helmet, and ground truth obtained with a terrestrial laser scanner. The experimental results indicate that our SLAM system is capable of creating quality exploration maps of small indoor spaces, and handling the loop-closure problem. The accumulated drift without loop closure was on average 1.1% (0.35 m) over a 31-m-long acquisition trajectory. Moreover, the comparison results demonstrated that our flying microdrone provided a comparable performance to the multilayer LiDAR-based macrodrone, given the low deviation between the point clouds built by both drones. Approximately 85 % of the cloud-to-cloud distances were less than 10 cm.
A complete multitemporal landslide inventory, ideally updated after each major event, is essential for quantitative landslide hazard assessment. However, traditional mapping methods, which rely on manual interpretation of aerial photographs and intensive field surveys, are time consuming and not efficient for generating such event-based inventories. In this letter, a semi-automatic approach based on object-oriented change detection for landslide rapid mapping and using very high resolution optical images is introduced. The usefulness of this methodology is demonstrated on the Messina landslide event in southern Italy that occurred on October 1, 2009. The algorithm was first developed in a training area of Altolia and subsequently tested without modifications in an independent area of Itala. Correctly detected were 198 newly triggered landslides, with user accuracies of 81.8% for the number of landslides and 75.9% for the extent of landslides. The principal novelties of this letter are as follows: 1) a fully automatic problem-specified multiscale optimization for image segmentation and 2) a multitemporal analysis at object level with several systemized spectral and textural measurements.
Abstract. Quick post-disaster actions demand automated, rapid and detailed building damage assessment. Among the available technologies, post-event oblique airborne images have already shown their potential for this task. However, existing methods usually compensate the lack of pre-event information with aprioristic assumptions of building shapes and textures that can lead to uncertainties and misdetections. However, oblique images have been already captured over many cities of the world, and the exploitation of pre- and post-event data as inputs to damage assessment is readily feasible in urban areas. In this paper, we investigate the potential of multi-temporal oblique imagery for detailed damage assessment focusing on two methodologies: the first method aims at detecting severe structural damages related to geometrical deformation by combining the complementary information provided by photogrammetric point clouds and oblique images. The developed method detected 87% of damaged elements. The failed detections are due to varying noise levels within the point cloud which hindered the recognition of some structural elements. We observed, in general that the façade regions are very noisy in point clouds. To address this, we propose our second method which aims to detect damages to building façades using the oriented oblique images. The results show that the proposed methodology can effectively differentiate among the three proposed categories: collapsed/highly damaged, lower levels of damage and undamaged buildings, using a computationally light-weight approach. We describe the implementations of the above mentioned methods in detail and present the promising results achieved using multi-temporal oblique imagery over the city of L’Aquila (Italy).
Human detection in images using deep learning has been a popular research topic in recent years and has achieved remarkable performance. Training a human detection network is useful for first responders to search for trapped victims in debris after a disaster. In this paper, we focus on the detection of such victims using deep learning, and we find that state-of-the-art detection models pre-trained on the well-known COCO dataset fail to detect victims. This is because all the people in the training set are shown in photos of daily life or sports activities, while people in the debris after a disaster usually only have parts of their bodies exposed. In addition, because of the dust, the colors of their clothes or body parts are similar to those of the surrounding debris. Compared with collecting images of common objects, images of disaster victims are extremely difficult to obtain for training. Therefore, we propose a framework to generate harmonious composite images for training. We first paste body parts onto a debris background to generate composite victim images and then use a deep harmonization network to make the composite images look more harmonious. We select YOLOv5l as the most suitable model, and experiments show that using composite images for training improves the AP (average precision) by 19.4% (15.3%→34.7%). Furthermore, using the harmonious images is of great benefit to training a better victim detector, and the AP is further improved by 10.2% (34.7%→44.9%). This research is part of the EU project INGENIOUS. Our composite images and code are publicly available on our website.
'%3e%3cpath fill='%23012169' d='M0 0v30h60V0z'/%3e%3cpath stroke='%23fff' stroke-width='6' d='m0 0 60 30m0-30L0 30'/%3e%3cpath stroke='%23C8102E' stroke-width='4' d='m0 0 60 30m0-30L0 30' clip-path='url(%23b)'/%3e%3cpath stroke='%23fff' stroke-width='10' d='M30 0v30M0 15h60'/%3e%3cpath stroke='%23C8102E' stroke-width='6' d='M30 0v30M0 15h60'/%3e%3c/g%3e%3c/svg%3e)
'%3e%3ccircle r='20' fill='%23008'/%3e%3ccircle r='17.5' fill='%23fff'/%3e%3ccircle r='3.5' fill='%23008'/%3e%3cg id='d'%3e%3cg id='c'%3e%3cg id='b'%3e%3cg id='a' fill='%23008'%3e%3ccircle r='.875' transform='rotate(7.5 -8.75 133.5)'/%3e%3cpath d='M0 17.5.6 7 0 2l-.6 5L0 17.5z'/%3e%3c/g%3e%3cuse xlink:href='%23a' transform='rotate(15)'/%3e%3c/g%3e%3cuse xlink:href='%23b' transform='rotate(30)'/%3e%3c/g%3e%3cuse xlink:href='%23c' transform='rotate(60)'/%3e%3c/g%3e%3cuse xlink:href='%23d' transform='rotate(120)'/%3e%3cuse xlink:href='%23d' transform='rotate(-120)'/%3e%3c/g%3e%3c/svg%3e)