With the popularization and wide application of drones in military and civilian fields, the safety of drones must be considered. At present, the failure and drop rates of drones are still much higher than those of manned aircraft. Therefore, it is imperative to improve the research on the safe landing and recovery of drones. However, most drone navigation methods rely on global positioning system (GPS) signals. When GPS signals are missing, these drones cannot land or recover properly. In fact, with the help of optical equipment and image recognition technology, the position and posture of the drone in three dimensions can be obtained, and the environment where the drone is located can be perceived. This paper proposes and implements a monocular vision-based drone autonomous landing system in emergencies and in unstructured environments. In this system, a novel map representation approach is proposed that combines three-dimensional features and a mid-pass filter to remove noise and construct a grid map with different heights. In addition, a region segmentation is presented to detect the edges of different-height grid areas for the sake of improving the speed and accuracy of the subsequent landing area selection. As a visual landing technology, this paper evaluates the proposed algorithm in two tasks: scene reconstruction integrity and landing location security. In these tasks, firstly, a drone scans the scene and acquires key frames in the monocular visual simultaneous localization and mapping (SLAM) system in order to estimate the pose of the drone and to create a three-dimensional point cloud map. Then, the filtered three-dimensional point cloud map is converted into a grid map. The grid map is further divided into different regions to select the appropriate landing zone. Thus, it can carry out autonomous route planning. Finally, when it stops upon the landing field, it will start the descent mode near the landing area. Experiments in multiple sets of real scenes show that the environmental awareness and the landing area selection have high robustness and real-time performance.

Vision-based mobile robot navigation is a vibrant area of research with numerous algorithms having been developed, the vast majority of which either belong to the scene-oriented simultaneous localization and mapping (SLAM) or fall into the category of robot-oriented lane-detection/trajectory tracking. These methods suffer from high computational cost and require stringent labelling and calibration efforts. To address these challenges, this paper proposes a lightweight robot navigation framework based purely on uncalibrated spherical images. To simplify the orientation estimation, path prediction and improve computational efficiency, the navigation problem is decomposed into a series of classification tasks. To mitigate the adverse effects of insufficient negative samples in the “navigation via classification” task, we introduce the spherical camera for scene capturing, which enables 360° fisheye panorama as training samples and generation of sufficient positive and negative heading directions. The classification is implemented as an end-to-end Convolutional Neural Network (CNN), trained on our proposed Spherical-Navi image dataset, whose category labels can be efficiently collected. This CNN is capable of predicting potential path directions with high confidence levels based on a single, uncalibrated spherical image. Experimental results demonstrate that the proposed framework outperforms competing ones in realistic applications.

Infrared human action recognition has many advantages, i.e., it is insensitive to illumination change, appearance variability, and shadows. Existing methods for infrared action recognition are either based on spatial or local temporal information, however, the global temporal information, which can better describe the movements of body parts across the whole video, is not considered. In this letter, we propose a novel global temporal representation named optical-flow stacked difference image (OFSDI) and extract robust and discriminative feature from the infrared action data by considering the local, global, and spatial temporal information together. Due to the small size of the infrared action dataset, we first apply convolutional neural networks on local, spatial, and global temporal stream respectively to obtain efficient convolutional feature maps from the raw data rather than train a classifier directly. Then these convolutional feature maps are aggregated into effective descriptors named three-stream trajectory-pooled deep-convolutional descriptors by trajectory-constrained pooling. Furthermore, we improve the robustness of these features by using the locality-constrained linear coding (LLC) method. With these features, a linear support vector machine (SVM) is adopted to classify the action data in our scheme. We conduct the experiments on infrared action recognition datasets InfAR and NTU RGB+D. The experimental results show that the proposed approach outperforms the representative state-of-the-art handcrafted features and deep learning features based methods for the infrared action recognition.

Vehicle surveillance of a wide area allows us to learn much about the daily activities and traffic information. With the rapid development of remote sensing, satellite video has become an important data source for vehicle detection, which provides a broader field of surveillance. The achieved work generally focuses on aerial video with moderately-sized objects based on feature extraction. However, the moving vehicles in satellite video imagery range from just a few pixels to dozens of pixels and exhibit low contrast with respect to the background, which makes it hard to get available appearance or shape information. In this paper, we look into the problem of moving vehicle detection in satellite imagery. To the best of our knowledge, it is the first time to deal with moving vehicle detection from satellite videos. Our approach consists of two stages: first, through foreground motion segmentation and trajectory accumulation, the scene motion heat map is dynamically built. Following this, a novel saliency based background model which intensifies moving objects is presented to segment the vehicles in the hot regions. Qualitative and quantitative experiments on sequence from a recent Skybox satellite video dataset demonstrates that our approach achieves a high detection rate and low false alarm simultaneously.

With the rapid development of Unmanned Aerial Vehicle (UAV) systems, the autonomous landing of a UAV on a moving Unmanned Ground Vehicle (UGV) has received extensive attention as a key technology. At present, this technology is confronted with such problems as operating in GPS-denied environments, a low accuracy of target location, the poor precision of the relative motion estimation, delayed control responses, slow processing speeds, and poor stability. To address these issues, we present a hybrid camera array-based autonomous landing UAV that can land on a moving UGV in a GPS-denied environment. We first built a UAV autonomous landing system with a hybrid camera array comprising a fisheye lens camera and a stereo camera. Then, we integrated a wide Field of View (FOV) and depth imaging for locating the UGV accurately. In addition, we employed a state estimation algorithm based on motion compensation for establishing the motion state of the ground moving UGV, including its actual motion direction and speed. Thereafter, according to the characteristics of the designed system, we derived a nonlinear controller based on the UGV motion state to ensure that the UGV and UAV maintain the same motion state, which allows autonomous landing. Finally, to evaluate the performance of the proposed system, we carried out a large number of simulations in AirSim and conducted real-world experiments. Through the qualitative and quantitative analyses of the experimental results, as well as the analysis of the time performance, we verified that the autonomous landing performance of the system in the GPS-denied environment is effective and robust.

Recognizing human actions from varied views is challenging due to huge appearance variations in different views. The key to this problem is to learn discriminant view-invariant representations generalizing well across views. In this paper, we address this problem by learning view-invariant representations hierarchically using a novel method, referred to as Joint Sparse Representation and Distribution Adaptation (JSRDA). To obtain robust and informative feature representations, we first incorporate a sample-affinity matrix into the marginalized stacked denoising Autoencoder (mSDA) to obtain shared features, which are then combined with the private features. In order to make the feature representations of videos across views transferable, we then learn a transferable dictionary pair simultaneously from pairs of videos taken at different views to encourage each action video across views to have the same sparse representation. However, the distribution difference across views still exists because a unified subspace where the sparse representations of one action across views are the same may not exist when the view difference is large. Therefore, we propose a novel unsupervised distribution adaptation method that learns a set of projections that project the source and target views data into respective low-dimensional subspaces where the marginal and conditional distribution differences are reduced simultaneously. Therefore, the finally learned feature representation is view-invariant and robust for substantial distribution difference across views even the view difference is large. Experimental results on four multiview datasets show that our approach outperforms the state-ofthe- art approaches.