Vehicular edge computing (VEC) is considered to be a key technology to improve the processing efficiency of computing tasks for the Internet of Vehicles (IoV). Using roadside units (RSUs) distributed on both sides of a road as edge servers, computation-intensive and latency-sensitive in-vehicle tasks can be responded to quickly. However, some Quality of Service (QoS) is often difficult to ensure due to clogged dense urban buildings or lack of infrastructure in remote areas. In this article, we propose a software-defined network (SDN)-driven partial offloading model for unmanned aerial vehicle (UAV)-assisted VEC networks, where the RSUs and UAVs jointly provide computing services to the vehicles and collect global information through centralized control using an SDN controller. To guarantee these vehicles obtain computing results in time and rationally utilize computing resources, we develop an optimal offloading mechanism using Age of Information (AoI), together with energy consumption and rental price as a comprehensive weighted cost of our above optimization objective. The total system cost of the performing tasks is minimized by jointly optimizing the UAV trajectory, user association, and offloading decision. Considering the mobility of the vehicles and UAVs and the dynamic network environment, we design a deep reinforcement learning (DRL)-based joint trajectory control and offloading allocation algorithm (DRL-TCOA) to solve the proposed computation offloading problem. Experimental results show that the proposed DRL-TCOA algorithm maintains better information freshness and lower system cost than the other baseline offloading strategies.

Positioning is envisioned as an essential enabler of future fifth generation (5G) mobile networks due to the massive number of use cases that would benefit from knowing users’ positions. In this work, we propose a particle filter-based reinforcement learning (PFRL) approach for the robust wireless indoor positioning system. Our algorithm integrates information of indoor zone prediction, inertial measurement units, wireless radio-based ranging, and floor plan into an particle filter. The zone prediction method is designed with an ensemble learning algorithm by integrating individual discriminative learning methods and Hidden Markov Models. Further, we integrate the particle filter approach with a reinforcement learning-based resampling method to provide robustness against localization failure problems such as the kidnapping robot problem. The PFRL approach is validated on a two-tier architecture, in which distributed machine learning tasks are hosted at client and edge layer. Experiment results show that our system outperforms traditional terminal-based approaches in both stability and accuracy.

Recently, multi-floor indoor positioning has become increasingly interesting for researchers, in which accurate recognition of indoor activities is critical for the detection of floor changes and the improvement of positioning accuracy according to indoor landmarks. However, we have not found a comprehensive study for recognizing indoor activities related to multi-floor indoor positioning based on a robust machine learning algorithm. In this work, we propose a framework for recognizing five indoor activities, i.e., walking, stillness, stair climbing, escalator or elevator taking. In this framework, we investigate the relevant sensors and features to improve the recognition accuracy of these activities, especially some specific features in frequency domain and wavelet domain. We propose to utilize a promising tree-based ensemble learning classifier, XGBoost, to recognize these activities. Based on our dataset created by 40 volunteers, we provide a comprehensive analysis of the proposed framework for indoor activity recognition. Considering both accuracy and computational cost, the XGBoost-based indoor activity recognition algorithm outperforms the other ensemble learning classifiers and single classifiers, and the average recognition F-score of XGBoost reaches 84.41%. Additionally, our introduced specific features in frequency domain and wavelet domain can significantly improve the recognition accuracy. Moreover, we use a publicly available dataset to verify our proposed framework and XGBoost classifier reaches 84.19% that outperforms the other classifiers.

Traditional WiFi positioning relies on a predefined radio map, which is labor-intensive and time-consuming for professionals. Recently, crowdsourcing has emerged as a promising solution for facilitating WiFi positioning. To crowdsense a radio map, traces collected from normal users are merged to recover the original walking paths. In this work, we design a robust iterative trace merging algorithm called WiFi-RITA based on WiFi access points as signal-marks. The algorithm formulates the trace merging problem as an optimization problem in which each trace is translated and rotated to minimize the limitation of distances among traces defined by WiFi access points. WiFi-RITA is further enhanced by removing outliers. WiFi-RITA is robust to the rotation errors of traces and efficient for a large number of short traces. According to the crowdsensed radio map, a sensor fusion approach based on particle filter by fusing inertial sensors and a multivariate Gaussian fingerprinting is proposed to enhance the accuracy of crowdsourcing indoor positioning. The experiment results in two large-scale environments demonstrate that WiFi-RITA positioning with zero-effort calibration achieves high positioning accuracy, which outperforms Pedestrian Dead Reckoning (PDR) and fingerprinting with K Nearest Neighbor.

Gesture recognition has been a hot research topic in human-computer interaction, since contactless gesture recognition will provide increasing applications in many fields. Millimeter-wave (mmWave) radar well serves this technology because of its high accuracy, easy integration, and strong anti-jamming ability in moving object detection. However, it is still challenging to meet the requirement of high precision in subtle gesture recognition based on traditional methods via point cloud or Range-Doppler heat map of millimeter-wave radar. Considering the raw data from millimeter-wave radar with more information such as phase, we propose a system that uses the constructed millimeter-wave radar data cube sequence and Timedistributed-CNN-Transformer network (CTN), called DCS-CTN system, to get higher hand gesture recognition accuracy. In this system, we introduce a time-distributed wrapper (TD) and convolution neural network (CNN) to extract local features of the data cube sequence, a position encoder to retain time information of the sequence, and a transformer network to get global features of the sequence. The experiments results show that this system can achieve hand gesture recognition accuracy of 99.75%, which is significantly higher than the other traditional approaches.

Millimeter-wave radar has demonstrated its high efficiency in complex environments in recent years, which outperforms LiDAR and computer vision in human activity recognition in the presence of smoke, fog, and dust. In previous studies, researchers mostly analyzed either 2D (3D) point cloud or range–Doppler information from radar echo to extract activity features. In this paper, we propose a multi-model deep learning approach to fuse the features of both point clouds and range–Doppler for classifying six activities, i.e., boxing, jumping, squatting, walking, circling, and high-knee lifting, based on a millimeter-wave radar. We adopt a CNN–LSTM model to extract the time-serial features from point clouds and a CNN model to obtain the features from range–Doppler. Then we fuse the two features and input the fused feature into the full connected layer for classification. We built a dataset based on a 3D millimeter-wave radar from 17 volunteers. The evaluation result based on the dataset shows that this method has higher accuracy than utilizing the two kinds of information separately and achieves a recognition accuracy of 97.26%, which is about 1% higher than other networks with only one kind of data as input.