Visual Place Recognition (VPR) is a critical technology for achieving robust long-term visual geo-localization. During the past few years, VPR research mainly focused on ground-based platforms in the street-level captured scenes with deep learning methods (e.g. NetVLAD, GeM), but little attention was paid to the VPR task on aerial vehicles. The algorithms and models designed for ground-based platforms are always directly applied to the aerial VPR problem. However, the viewpoint variance on Unmanned Aerial Vehicles (UAV) is much larger than the ground-based platforms. Due to the sparse distribution of aerial image features, when the viewpoint of the camera changes, the features of the query image are largely inconsistent with the descriptors in the database, which results in the failures of image retrieval and visual geo-localization. In this letter, we propose an aerial VPR enhancement module called GeoCluster , which presents a feature aggregation method using spatial clustering information to improve the robustness and consistency of the global descriptors for UAV-captured frames. Moreover, it can be applied to any NetVLAD-based VPR method and boost the pre-trained model without any further training process. By integrating GeoCluster into an existing state-of-the-art localization method, we can achieve about 10% improvement for aerial image retrieval tasks and have more accurate and robust geo-localization results.

This paper presents an efficient and accurate range-aided visual-inertial odometry (RVIO) system for the global positioning system denied environment. In particular, the ultra-wideband (UWB) measurements are integrated to reduce the long-term drift of the visual-inertial odometry (VIO) system. Our approach starts with a filter-based scheme to localize the unknown UWB anchor in the local world frame. In particular, a novel surface-based particle filter is proposed to localize the UWB anchors efficiently. When the initialization is complete, the UWB location information is utilized to support the subsequent long-term robot positioning. An observability-constrained optimization approach is developed to combine the visual, inertial, and UWB range measurements. Such a framework takes advantage of both VIO and UWB measurements and is feasible even when the number of observed UWB anchors is below four. Experiments on both simulated and real-world scenes demonstrate the validity and superiority of the proposed system.

Multi-robot SLAM (simultaneous localization and mapping) is crucial for the implementation of robots in practical scenarios. Bandwidth constraints significantly influence multi-robot SLAM systems, prompting a reliance on lightweight feature descriptors for robot cooperation in positioning tasks. Real-time map sharing among robots is also frequently ignored in such systems. Consequently, such algorithms are not feasible for autonomous multi-robot navigation tasks in the real world. Furthermore, the computation cost of the global optimization of multi-robot SLAM increases significantly in large-scale scenes. In this study, we introduce a novel distributed multi-robot SLAM framework incorporating sliding window-based optimization to mitigate computation loads and manage inter-robot loop closure constraints. In particular, we transmit a 2.5D grid map of the keyframe-based submap between robots to promote map consistency among robots and maintain bandwidth efficiency in data exchange. The proposed algorithm was evaluated in extensive experimental environments, and the results validate its effectiveness and superiority over other mainstream methods.

In this work, we propose a 3D occlusion facial recognition network based on a multi-feature combination threshold (MFCT-3DOFRNet). First, we design and extract the depth information of the 3D face point cloud, the elevation, and the azimuth angle of the normal vector as new 3D facially distinctive features, so as to improve the differentiation between 3D faces. Next, we propose a multi-feature combinatorial threshold that will be embedded at the input of the backbone network to implement the removal of occlusion features in each channel image. To enhance the feature extraction capability of the neural network for missing faces, we also introduce a missing face data generation method that enhances the training samples of the network. Finally, we use a Focal-ArcFace loss function to increase the inter-class decision boundaries and improve network performance during the training process. The experimental results show that the method has excellent recognition performance for unoccluded faces and also effectively improves the performance of 3D occlusion face recognition. The average Top-1 recognition rate of the proposed MFCT-3DOFRNet for the Bosphorus database is 99.52%, including 98.94% for occluded faces and 100% for unoccluded faces. For the UMB-DB dataset, the average Top-1 recognition rate is 95.08%, including 93.41% for occluded faces and 100% for unoccluded faces. These 3D face recognition experiments show that the proposed method essentially meets the requirements of high accuracy and good robustness.

The lack of a sufficient number of reliable corners in low-textured environments is a big challenge for classical visual Simultaneous Localization And Mapping (SLAM), especially for point feature-based methods. Many other features (i.e., line and plane segments) are often combined with points to restore an environmental structure. However, using such features requires much computational time. This work focuses on the reliable and high-performance real-time operation of an SLAM system in low-textured scenarios. It proposes a semidirect multimap monocular SLAM system (SM-SLAM) that combines direct tracking and feature-based map maintenance with point features and line segments. The proposed system tracks nonkeyframes based on a sparse image alignment method for fast tracking, extracts and matches point features and line segments in keyframes for high-quality environment structure and motion optimization. We present an extensive evaluation on two widely-used datasets and some challenging real-world scenarios. Experimental results show that SM-SLAM can well reconstruct a sparse 3-D map with geometrical structure information in 30–40 Hz. It shows an accuracy improvement of more than 20% than the Oriented FAST and Rotated BRIEF feature-based SLAM on some low-speed, small-range camera motion datasets and performs well in low-texture scenarios.

Effective training of the deep neural networks requires much data to avoid underdetermined and poor generalization. Data Augmentation alleviates this by using existing data more effectively. However standard data augmentation produces only limited plausible alternative data by for example, flipping, distorting, adding noise to, cropping a patch from the original samples. In this paper, we introduce the adversarial autoencoder (AAE) to impose the feature representations with uniform distribution and apply the linear interpolation on latent space, which is potential to generate a much broader set of augmentations for image classification. As a possible "recognition via generation" framework, it has potentials for several other classification tasks. Our experiments on the ILSVRC 2012, CIFAR-10 datasets show that the latent space interpolation (LSI) improves the generalization and performance of state-of-the-art deep neural networks.

In visual tracking, subspace representation has been proven effective in many cases. However, in some scenarios the target observations may contain outliers (e.g. pixels that are occluded) for which subspace representation does not hold. In order to deal with this issue and noticing that outliers usually occur in clusters and occupy only a fraction of the whole target image patch, we describe them by a block sparse vector and add it to the original subspace model. Under the framework of variational Bayesian inference, we then develop an algorithm to recover the sparse vector. With the knowledge of such a vector, the outliers can be detected and removed and subspace representation can be conducted over the outlier-free observations. Based on this strategy, the proposed tracking method needs to perform sparse recovery only once for each frame, which brings significant computational savings compared to existing tracking methods that exploit sparse representation. Moreover, unlike the existing methods that typically employ the ℓ 1 norm regularization approach to achieve sparse recovery and have to set the regularization parameter appropriately, the proposed method can estimate the unknown coefficient vectors and all other unknown model parameters in an automatic manner and thus requires less user intervention in the tracking process. Experimental results on various videos demonstrate that the method can achieve good tracking performance at a low computational cost.

Extracting road networks from very-high-resolution VHR aerial and satellite imagery has been a long-standing problem. In this article, a neural-dynamic tracking framework is proposed to extract road networks based on deep convolutional neural networks DNN and a finite state machine FSM. Inspired by autonomous mobile systems, the authors train a DNN to recognize the pattern of input data, which is an image patch extracted in a detection window centred at the current location of the tracker. The pattern is predefined according to the environment and associated with the states in the FSM. A vector-guided sampling method is proposed to generate the training data set for the DNN, which extracts massive image-direction pairs from the imagery and existing vector road maps. In the tracking procedure, the size of the detection window is determined by a fusion strategy and the extracted image patches represent the orientation features of the road local environment that can be recognized by the trained DNN. The reactive unit in FSM associates states with behaviours of the tracker while continually modifying the orientation to follow the road and generating a sequence of states and locations. In this way, our framework combines the DNN and FSM. DNN acts as a key component to recognize patterns from a complex and changing environment; FSM translates the recognized patterns to states and controls the behaviour of the tracker. The results illustrate that our approach is more accurate and efficient than the traditional ones.

The signal-to-noise ratio (SNR) and spatial resolution are conflicting factors in conventional Brillouin optical time domain reflectometer (BOTDR) system. Golay complementary sequence with good autocorrelation properties, can be used as probe pulse, in order to achieve a higher SNR without compromising the spatial resolution of the system. This paper discusses the composition method of Golay complementary sequences, studies the principle applying Golay complementary sequences in BOTDR system. On this basis, a heterodyne detection BOTDR sensing system using Golay complementary sequences is designed and the system performances in time and frequency domain are simulated numerically. The results show that the method can improve the SNR of the system without degrading the spatial resolution.

A high-performance lossy compression algorithm was proposed for remote sensing image compression based on bit allocation using sub-bands entropy.After decomposing the remote sensing image by using wavelet lifting scheme,the distribution of energy percentage and entropy of high frequency sub-bands were analyzed.A novel bit allocation method using the entropy was proposed.Uniform scalar quantification was implemented for each high frequency sub-band.Bit plane encoding followed and included two parts.The coordinates of non-zero coefficients were registered in the most significant bit plane,and run-length encoding and Huffman encoding were adopted for other bit planes. Experimental results show that the compression scheme performs well on a set of test remote sensing images including complicated texture images and relative planar ones. The Peak Signal to Noise Ratio(PSNR)is all higher than 34dB.However,the Compression Ratio(CR)depends on image's complicated degree.

A simple and adaptive lossless compression algorithm is proposed for remote sensing image compression, which includes integer wavelet transform and the Rice entropy coder. By analyzing the probability distribution of integer wavelet transform coefficients and the characteristics of Rice entropy coder, the divide and rule method is used for high-frequency sub-bands and low-frequency one. High-frequency sub-bands are coded by the Rice entropy coder, and low-frequency coefficients are predicted before coding. The role of predictor is to map the low-frequency coefficients into symbols suitable for the entropy coding. Experimental results show that the average Compression Ratio (CR) of our approach is about two, which is close to that of JPEG 2000. The algorithm is simple and easy to be implemented in hardware. Moreover, it has the merits of adaptability, and independent data packet. So the algorithm can adapt to space lossless compression applications.