For image semantic segmentation, a fully convolutional network is usually employed as the encoder to abstract visual features of the input image. A meticulously designed decoder is used to decoding the final feature map of the backbone. The output resolution of backbones which are designed for image classification task is too low to match segmentation task. Most existing methods for obtaining the final high-resolution feature map can not fully utilize the information of different layers of the backbone. To adequately extract the information of a single layer, the multi-scale context information of different layers, and the global information of backbone, we present a new attention-augmented module named Dense-attention Context Module (DCM), which is used to connect the common backbones and the other decoding heads. The experiments show the promising results of our method on Cityscapes dataset.

Frequency-modulated continuous-wave (FMCW) is widely used in radar systems to measure range and speed due to its low cost, outstanding reliability and full automation. In this paper, a novel framework is proposed to enable the communication functionality of millimeter wave (mmWave) FMCW radar. The symbol is modulated by the frequency slope of chirp pulse, when matched filter based approach is used to demodulate the information symbol. Theoretical analysis is provided for the upper bound of the symbol error rate by introducing the Fresnel integral. Based on the measurements of our mmWave FMCW radar test bed, a detailed simulation is carried out to demonstrate the performance of the proposed framework.

Millimeter wave (mmWave) provides an enormous spectrum for future broadband cellular communications. The corresponding challenging characteristics of radio propagation, and the use of highly directional transmission and the dense deployment, lead to more complex and critical resource allocation problems than in traditional cellular systems, where the channels are better tamed. In this paper, we study the joint power control and user association problem in heterogeneous networks (HetNets) by considering the dynamics of links as a Markov Decision Process (MDP). A reinforcement learning framework is proposed to study the problem. The large state/action space is handled by decomposing the large scale problem into multiple local problems, based on the topology of mmWave HetNet, which is motivated by the celebrated belief propagation (BP) algorithm in probabilistic graphical models. The decomposed problem is solved with a distributed message passing method and accelerated by the prior knowledge of the mmWave dynamics. Two categories of learning frameworks are proposed for time sensitive and power sensitive conditions. The real-world measurements in the 60GHz band are collected and used in the simulation of the proposed framework.

Millimeter wave is a promising technique in the next generation of cellular communications due to its large bandwidth and thus unprecedentedly high data transmission rate. In this paper, the blocking effect of millimeter wave is studied through a series of experiments. Then, the blocking of millimeter wave communications is applied to track the movement of the blocking object. Two object tracking schemes via the blocking effect, using two different criteria, are proposed. A walking trajectory model is proposed and analyzed to evaluate the performance of the proposed tracking algorithm analytically. The performance of the proposed schemes is tested by simulations on true trajectories based on GPS measurement.

A prominent feature of millimeter wave communications is the blockage effect: an object intruding the line-of-sight (LOS) of a communication link can block the millimeter wave signal. While the blockage incurs outage to the communications, it also brings information related to the blocking object. In this paper, both short-term and long- term trajectory estimations are studied for the object(s) blocking the millimeter wave communication link(s). For the short-term trajectory, experiment has revealed signal strength oscillations before and after the blockage, due to the reflection at the blocking object and the interference with LOS signal. It is shown that the short-term trajectory, as well as the speed, of the moving object can be estimated from the samples of oscillations. For the long-term trajectory, the tracking of multiple objects is carried out using the blockage events on multiple millimeter wave communication links. The trajectory estimation is converted to the problem of object recognition, which is solved by exploiting the geometric intuition. Numerical and experimental results show that the proposed short-term and long-term trajectory estimations achieve excellent performances.

This paper proposes a distributed multiagent reinforcement learning framework for resource allocation problems in heterogeneous wireless networks (HetNet). The base station, pico transmitters and user equipments collaborate to improve the system throughput and energy efficiency by handling the inter-cell interference based on the local observable environment. Utilizing the sparsity and tree structure of connection graph of the network, the collaboration is carried out through a distributed calculation of the optimal action via the message passing method. The tradeoff between the system throughput and the energy efficiency is addressed by a weighting parameter in the reward function. Simulation results show that the proposed framework outperforms the traditional inter-cell interference coordination method and improves overall performance.

Communication is a bottleneck for many distributed machine learning problems with large data sets. One effective approach is to exchange the set of good classifiers among the learners. The key challenge in such an approach is how to represent the set of good classifiers, which can be viewed as a source coding problem in communication systems. A nonuniform sampling framework is proposed for this source coding problem. A metric that describes the quality of the sampled data for the target hypothesis space is proposed. An optimization problem is formulated by minimizing the upper bound of the proposed metric, and an efficient weight based algorithm is provided to solve the optimization problem. Numerical simulations on synthetic and real world data set show that the learning performance based on the proposed sampling framework is close to the global optimum with high confidence, while requiring less samples than baseline random sampling algorithms.

Over the past decades, a wide attention has been paid to crowd control and management in the intelligent video surveillance area. Among the tasks for automatic surveillance video analysis, crowd motion modeling lays a crucial foundation for numerous subsequent analysis but encounters many unsolved challenges due to occlusions among pedestrians, complicated motion patterns in crowded scenarios, etc. Addressing the unsolved challenges, the authors propose a novel spatio-temporal viscous fluid field to model crowd motion patterns by exploring both appearance of crowd behaviors and interaction among pedestrians. Large-scale crowd events are hereby recognized based on characteristics of the fluid field. First, a spatio-temporal variation matrix is proposed to measure the local fluctuation of video signals in both spatial and temporal domains. After that, eigenvalue analysis is applied on the matrix to extract the principal fluctuations resulting in an abstract fluid field. Interaction force is then explored based on shear force in viscous fluid, incorporating with the fluctuations to characterize motion properties of a crowd. The authors then construct a codebook by clustering neighboring pixels with similar spatio-temporal features, and consequently, crowd behaviors are recognized using the latent Dirichlet allocation model. The convincing results obtained from the experiments on published datasets demonstrate that the proposed method obtains high-quality results for large-scale crowd behavior perception in terms of both robustness and effectiveness.

Over the past decades, a wide attention has been paid to crowd control and management in intelligent video surveillance area. In this paper, the authors propose a novel spatiotemporal viscous fluid field to recognize large-scale crowd event with respect to both appearance and driven factor of crowd behavior. Firstly, a spatiotemporal variation matrix is proposed to exploit motion property of a crowd. In particular, the paper exploits characteristics of the matrix with eigenvalue decomposition algorithm and constructs an abstract fluid field to model the crowd motion pattern, which is denoted by spatiotemporal fluid field. Secondly, the paper proposes a spatiotemporal force field to exploit the interaction force between the pedestrians. Furthermore, the fluid and force field constructs a spatiotemporal viscous fluid field. Thirdly, after generating feature with bag of word model, the authors utilize latent Dirichlet allocation model to recognize crowd behavior. The experiments on PETS2009 and UMN datasets show that the proposed method has a better performance for large-scale crowd behavior perception in both robustness and effectiveness comparing with the conventional methods.

Motion analysis is still a challenging task in many computer vision applications. This paper proposes a new low-level motion feature based on geometric regularity information, particularly suited for traffic surveillance. Firstly, a novel concept of temporal geometry consistency constraint (TGCC) is introduced, which exploits the fact that the geometric structure of a rigid object remains consistent across consecutive frames. Furthermore, the spatial geometric flow is adopted to characterize image structure. Finally, the video motion is represented as a set of geometric flow that moves in the temporal direction. In this case, the method yields a promising illumination robust moderately dense geometric motion flow (GMF) and has more explicit motion boundaries. The GMF could also be used for higher level motion modeling and structural inference tasks, as an effective low-level feature. Extensive experiment results on real video demonstrate the effectiveness and robustness of the proposed method for vehicle motion analysis.

Over the past decade, a wide attention has been paid to the crowd control and management in intelligent video surveillance area. This paper proposes a sparse spatiotemporal local binary pattern (SST-LBP) descriptor to extract the dynamic texture of the walking crowd with the application to crowd density estimation. Firstly, the sparse selected location is extracted, which is notably variant in temporal domain and scale invariant in spatial domain. Afterwards, considering the spatial and temporal symmetry, the authors propose a sparse spatiotemporal local binary pattern algorithm and utilize its statistical property to describe the crowd feature. Finally, the crowd features are classified into a range of density levels by adopting support vector machine. The experiments on real video show that the proposed SST-LBP method is effective and robust on the large-scale crowd density estimation. Compared with the other methods, the proposed method does not base on the premise that the background should be extracted perfectly, which is too complicated to implement in real surveillance.