We propose a framework for the integration of heterogeneous networks in human pose estimation (HPE) with the aim of balancing accuracy and computational complexity. Although many existing methods can improve the accuracy of HPE using multiple frames in videos, they also increase the computational complexity. The key difference here is that the proposed heterogeneous framework has various networks for different types of frames, while existing methods use the same networks for all frames. In particular, we propose to divide the video frames into two types, including key frames and non-key frames, and adopt three networks including slow networks, fast networks, and transfer networks in our heterogeneous framework. For key frames, a slow network is used that has high accuracy but high computational complexity. For non-key frames that follow a key frame, we propose to warp the heatmap of a slow network from a key frame via a transfer network and fuse it with a fast network that has low accuracy but low computational complexity. Furthermore, when extending to the usage of long-term frames where a large number of non-key frames follow a key frame, the temporal correlation decreases. Therefore, when necessary, we use an additional transfer network that warps the heatmap from a neighboring non-key frame. The experimental results on PoseTrack 2017 and PoseTrack 2018 datasets demonstrate that the proposed FSPose achieves a better balance between accuracy and computational complexity than the competitor method. Our source code is available at https://github.com/Fenax79/fspose.

This paper proposes a low-complexity variational Bayesian inference (VBI)-based method for massive multiple-input multiple-output (MIMO) downlink channel estimation. The temporal correlation at the mobile user side is jointly exploited to enhance the channel estimation performance. The key to the success of the proposed method is the column-independent factorization imposed in the VBI framework. Since we separate the Bayesian inference for each column vector of signal-of-interest, the computational complexity of the proposed method is significantly reduced. Moreover, the temporal correlation is automatically uncoupled to facilitate the updating rule derivation for the temporal correlation itself. Simulation results illustrate the substantial performance improvement achieved by the proposed method.

The large computational complexity makes stereo matching a big challenge in real-time application scenario. The problem of stereo matching in a video sequence is slightly different with that in a still image because there exists temporal correlation among video frames. However, no existing method considered temporal consistency of disparity for algorithm acceleration. In this work, we proposed a scheme called the dynamic disparity range (DDR) to optimize matching cost calculation and cost aggregation steps by narrowing disparity searching range, and a scheme called temporal cost aggregation path to optimize the cost aggregation step. Based on the schemes, we proposed the DDR-SGM and the DDR-MCCNN algorithms for the stereo matching in video sequences. Evaluation results showed that the proposed algorithms significantly reduced the computational complexity with only very slight loss of accuracy. We proved that the proposed optimizations for the stereo matching are effective and the temporal consistency in stereo video is highly useful for either improving accuracy or reducing computational complexity.

Energy harvesting technology was introduced into wireless sensor networks (WSNs) to solve the problem of the short lifetimes of sensor nodes. The technology gives sensor nodes the ability to convert environmental energy into electricity. Sufficient electrical energy can lengthen the lifetime and improve the quality of service of a WSN. This paper proposes a novel use of mutual information to evaluate data transmission behavior in the energy harvesting WSNs. Data at a sink for a node deteriorates over time until the next periodic transmission from the node is received. In this paper, we suggest an optimized intermittent transmission method for WSNs that harvest energy. Our method overcomes the problem of information deterioration without increasing energy cost. We show that by using spatial correlation between different sensor nodes, our proposed method can mitigate information deterioration significantly at the sink.

A stereo video coding scheme which is compatible with monoview-processor is presented in this paper. At the same time, this paper proposes an adaptive prediction structure which can make different prediction modes to be applied to different groups of picture (GOPs) according to temporal correlations and interview correlations to improve the coding efficiency. Moreover, the most advanced video coding standard H.264 is used conveniently for maximize the coding efficiency in this paper. Finally, the effectiveness of the proposed scheme is verified by extensive experimental results.

In adaptive motion estimation, spatial-temporal correlation based motion type inference has been recognized as an effective way to guide the motion estimation strategy adjustment according to video contents. However, the complexity and the reliability of those methods remain two crucial problems. In this paper, a motion vector field model is introduced as the basis for a new spatial-temporal correlation based motion type inference method. For each block, Full Search with Adaptive Search Window (ASW) and Three Step Search (TSS), as two search strategy candidates, can be employed alternatively. Simulation results show that the proposed method can constantly reduce the dynamic computational cost to as low as 3% to 4% of that of Full Search (FS), while remaining a closer approximation to FS in terms of visual quality than other fast algorithms for various video sequences. Due to its efficiency and reliability, this method is expected to be a favorable contribution to the mobile video communication where low power real-time video coding is necessary.

This paper proposes a new method for recovering the original signals from their linear mixtures observed by the same number of sensors. It is performed by identifying the linear transform from the sources to the sensors, only using the sensor signals. The only assumption of the source signals is basically the fact that they are statistically mutually independent. In order to perform the 'blind' identification, some time-correlational information in the observed signals are utilized. The most important feature of the method is that the full information of available time-correlation data (second-order statistics) is evaluated, as opposed to the conventional methods. To this end, an information-theoretic cost function is introduced, and the unknown linear transform is found by minimizing it. The propsed method gives a more stable solution than the conventional methods.

The convergence characteristics of the adaptive beamformer with the RLS algorithm are analyzed in this paper. In case of the RLS adaptive beamformer, the convergence characteristics are significantly affected by the spatial characteristics of the signals/noises in the environment. The purpose of this paper is to show how these physical parameters affect the convergence characteristics. In this paper, a typical environment where a few directional noises are accompanied by background noise is assumed, and the influence of each component of the environment is analyzed separately using rank analysis of the correlation matrix. For directional components, the convergence speed is faster for a smaller number of noise sources since the effective rank of the input correlation matrix is reduced. In the presence of background noise, the convergence speed is slowed down due to the increase of the effective rank. However, the convergence speed can be improved by controlling the initial matrix of the RLS algorithm. The latter section of this paper focuses on the physical interpretation of this initial matrix, in an attempt to elucidate the mechanism of the convergence characterisitics.