Pedestrian trajectory prediction is an essential task in robotic applications such as autonomous driving and robot navigation. State-of-the-art predictors use a conditional variational autoencoder (CVAE) with recurrent neural networks (RNNs) to encode observed trajectories decode multi-modal future trajectories. This process can suffer from accumulated errors over long horizons (≥2 seconds). letter presents BiTraP, goal-conditioned bi-directional method based on the CVAE. BiTraP estimates...

This paper introduces the beta compositional model (BCM) for hyperspectral unmixing and four algorithms given BCM. Hyperspectral estimates proportion of each endmember at every pixel a image. Under BCM, is random variable distributed according to distribution. By using distribution, spectral variability accounted during unmixing, reflectance values are constrained physically realistic range, skew can be in Spectral incorporated increase accuracy. Two BCM-based approaches presented:...

Accurate prediction of pedestrian crossing behaviors by autonomous vehicles can significantly improve traffic safety. Existing approaches often model using trajectories or poses but do not offer a deeper semantic interpretation person's actions how influence pedestrian's intention to cross in the future. In this work, we follow neuroscience and psychological literature define behavior as combination an unobserved inner will (a probabilistic representation binary intent vs. crossing) set...

This paper applies the rooting theory to study and analyze how short video “China Mosaic” constructs national image on Facebook. Through three levels of coding: open coding, spindle selective it summaries communication influencing factors in constructing presenting from five dimensions, such as politics, economy, culture, natural environment, society, with a view providing reference samples for mainstream media image.

In classifier (or regression) fusion the aim is to combine outputs of several algorithms boost overall performance. Standard supervised often require accurate and precise training labels. However, labels may be difficult obtain in many remote sensing applications. This paper proposes novel classification regression models that can trained given ambiguosly imprecisely labeled data which are associated with sets points (i.e., "bags") instead individual "instances") following a multiple...

In applications such as autonomous driving, it is important to understand, infer, and anticipate the intention future behavior of pedestrians. This ability allows vehicles avoid collisions improve ride safety quality. paper proposes a biomechanically inspired recurrent neural network (Bio-LSTM) that can predict location 3D articulated body pose pedestrians in global coordinate frame, given poses locations estimated prior frames with inaccuracy. The proposed able for multiple simultaneously,...

Pedestrian detection is an important task for human-robot interaction and autonomous driving applications. Most previous pedestrian methods rely on data collected from three-dimensional (3D) Light Detection Ranging (LiDAR) sensors in addition to camera imagery, which can be expensive deploy. In this letter, we propose a novel Planar LiDAR Pose Network (PPLP Net) based two-dimensional (2D) monocular offers far more affordable solution the oriented problem. The proposed PPLP Net consists of...

The Multiple Instance Choquet integral (MICI) for classifier fusion and an evolutionary algorithm parameter estimation is presented. has a long history of providing effective framework non-linear fusion. However, previous methods to learn appropriate measure the required accurate precise training labels. In many applications, data-point specific labels are unavailable infeasible obtain. proposed MICI allows with uncertain in which class provided sets data points (i.e., "bags") instead...

Urban environments pose a significant challenge for autonomous vehicles (AVs) as they must safely navigate while in close proximity to many pedestrians. It is crucial the AV correctly understand and predict future trajectories of pedestrians avoid collision plan safe path. Deep neural networks (DNNs) have shown promising results accurately predicting pedestrian trajectories, relying on large amounts annotated real-world data learn behavior. However, collecting annotating these datasets...

Classifier fusion methods integrate complementary information from multiple classifiers or detectors and can aid remote sensing applications such as target detection hyperspectral image analysis. The Choquet integral (CI), parameterized by fuzzy measures (FMs), has been widely used in the literature an effective non-linear framework. Standard supervised CI algorithms often require precise ground-truth labels for each training data point, which be difficult impossible to obtain data....

Sensor fusion combines data from multiple sensor sources to improve reliability, robustness, and accuracy of interpretation. The Fuzzy Integral (FI), in particular, the Choquet integral (ChI), is often used as a powerful nonlinear aggregator for across sensors. However, existing supervised ChI learning algorithms typically require precise training labels each input point, which can be difficult or impossible obtain. Additionally, prior work on based only normalized fuzzy measures, bounds...

This paper proposes a possibilistic context identification approach for synthetic aperture sonar (SAS) imagery. SAS seabed imagery can display variety of textures that be used to identify types such as sea grass, sand ripple and hard-packed sand, etc. Target objects in often have varying characteristics features due changing environmental context. Therefore, methods the environment assist target classification detection an environmentally adaptive or context-dependent approach. In this...

In remote sensing, each sensor can provide complementary or reinforcing information. It is valuable to fuse outputs from multiple sensors boost overall performance. Previous supervised fusion methods often require accurate labels for pixel in the training data. However, many remote-sensing applications, pixel-level are difficult infeasible obtain. addition, have different resolutions modalities. For example, rasterized hyperspectral imagery (HSI) presents data a grid while airborne light...

An approach to image labeling by seabed context based on multiple-instance learning via embedded instance selection (MILES) is presented. Sonar images are first segmented into superpixels with associated intensity and texture feature distributions. These defined as the "instances" sonar "bags" within MILES classification framework. The distributions discrete while continuous, thus Cauchy-Schwarz divergence metric used embed instances in a higher-dimensional discriminatory space. Results...

Video anomaly detection is a core problem in vision. Correctly detecting and identifying anomalous behaviors pedestrians from video data will enable safety-critical applications such as surveillance, activity monitoring, human-robot interaction. In this paper, we propose to leverage trajectory localization prediction for unsupervised pedestrian event detection. Different than previous reconstruction-based approaches, our proposed framework rely on the errors of normal abnormal trajectories...

Pedestrian pose prediction is an important topic, related closely to robotics and automation. Accurate predictions of human poses motion can facilitate a more thorough understanding analysis behavior, which benefits real-world applications such as human-robot interaction, humanoid bipedal robot design, safe navigation mobile robots autonomous vehicles. This article describes deep predictive coding network (PredNet)-based approach for unsupervised pedestrian from 2D camera imagery provides...

Pedestrian trajectory prediction is an essential task in robotic applications such as autonomous driving and robot navigation. State-of-the-art predictors use a conditional variational autoencoder (CVAE) with recurrent neural networks (RNNs) to encode observed trajectories decode multi-modal future trajectories. This process can suffer from accumulated errors over long horizons (>=2 seconds). paper presents BiTraP, goal-conditioned bi-directional method based on the CVAE. BiTraP estimates...

Characteristics of underwater targets displayed in synthetic aperture sonar (SAS) imagery vary depending on their environmental context. Discriminative features sea grass may differ from the that are discriminative sand ripple, for example. Environmentally-adaptive target detection and classification systems take into account context, therefore, have potential improved results. This paper presents an end-to-end environmentally-adaptive system SAS performs recognition while accounting First,...