Changzhen Li

Wireless channel analysis forms the foundation for characterizing channel models. Studying the channel characteristics in different scenarios is significant for reducing interference and enhancing channel stability. In this paper, Vehicle-to-Infrastructure (V2I) channel measurements were conducted in vegetated areas of Beijing and Wuhan at frequencies of 5.9 GHz and 5.8 GHz, respectively, and the dynamic channel characteristics in these two environments were compared. Additionally, a tapped delay line (TDL) model was used to analyze the multipath effects. The results indicate that higher environmental dynamics lead to shorter channel stationarity durations, with measured values of 51.72 ms in Wuhan scenario and 10 ms in Beijing scenario. The RMS delay spread (RMS-DS) is observed to be larger in the Wuhan scenario, with 90% of the values below 168.19 ns, while the corresponding value is 8.48 ns in the Beijing scenario. This paper provides a reference for research in similar environments by comparatively analyzing different channel characteristics and establishing channel models.

The detection of traffic signs through vehicle camera sensors is a crucial task for intelligent vehicles. Prohibition sign is one of the traffic signs that have a significant impact on traffic safety. However, previous research has rarely specifically focused on prohibition signs, and the accurate and rapid detection of prohibition signs remains a challenge. In this paper, a dataset with 12 most common types of Chinese prohibition signs is established, and an improved lightweight model specifically for prohibition sign detection is proposed based on YOLOv5n. In the early stage of the feature extraction network, Shuffle Attention (SA) is introduced to capture detailed features of the target. In the feature fusion network, a multi-path weighted fusion approach is employed to fully utilize valuable information from feature maps at different scales. In the detection output network, to prevent adverse effects on detection performance from other scales, only the 80x80 detection head scale is retained. Experimental results on the established dataset show that, compared to the original model, the detection accuracy mAP@0.5 of the improved model is increased by 20.4%, with a simultaneous 31.7% reduction in the parameter count. Compared with other lightweight models such as Faster RCNN, SSD, and YOLO series, the proposed model also has higher detection accuracy and fewer model parameters. Additionally, the model is deployed on Jetson TX2 and Raspberry Pi 4B with detection speeds of 21.3 and 1.8 FPS respectively, showcasing its potential for practical applications of vehicle camera sensors. The code of the proposed model is shared online at https://github.com/Tomming-Tong/yolov5_ms8.

Recently action recognition has received more and more attention for its comprehensive and practical applications in intelligent surveillance and human-computer interaction. However, few-shot action recognition has not been well explored and remains challenging because of data scarcity. In this paper, we propose a novel hierarchical compositional representations (HCR) learning approach for few-shot action recognition. Specifically, we divide a complicated action into several sub-actions by carefully designed hierarchical clustering and further decompose the sub-actions into more fine-grained spatially attentional sub-actions (SAS-actions). Although there exist large differences between base classes and novel classes, they can share similar patterns in sub-actions or SAS-actions. Furthermore, we adopt the Earth Mover's Distance in the transportation problem to measure the similarity between video samples in terms of sub-action representations. It computes the optimal matching flows between sub-actions as distance metric, which is favorable for comparing fine-grained patterns. Extensive experiments show our method achieves the state-of-the-art results on HMDB51, UCF101 and Kinetics datasets.

Vehicle detection is crucial for traffic surveillance and assisted driving. To overcome the loss of efficiency, accuracy, and stability in low-light conditions, we propose a lightweight “You Only Look Once” (YOLO) detection model. A polarized self-attention-enhanced aggregation feature pyramid network is used to improve feature extraction and fusion in low-light scenarios, and enhanced “Swift” spatial pyramid pooling is used to reduce model parameters and enhance real-time nighttime detection. To address imbalanced low-light samples, we integrate an anchor mechanism with a focal loss to improve network stability and accuracy. Ablation experiments show the superior accuracy and real-time performance of our Light-YOLO model. Compared with EfficientNetv2-YOLOv5, Light-YOLO boosts mAP@0.5 and mAP@0.5:0.95 by 4.03 and 2.36%, respectively, cuts parameters by 44.37%, and increases recognition speed by 20.42%. Light-YOLO competes effectively with advanced lightweight networks and offers a solution for efficient nighttime vehicle-detection.

Under the framework of sixth-generation (6G) wireless communications, the unmanned aerial vehicle (UAV) plays an irreplaceable role in a number of communication systems. In this paper, a novel cluster-based low-altitude UAV-to-vehicle (U2V) non-stationary channel model with uniform planar antenna arrays (UPAs) is proposed. In order to comprehensively model the scattering environment, both single and twin clusters are taken into account. A novel continuous cluster evolution algorithm that integrates time evolution and array evolution is developed to capture channel non-stationarity. In the proposed algorithm, the link between the time evolution of twin clusters and that of single clusters is established to regulate the temporal evolution trend. Moreover, an improved observable radius method is applied to UPAs for the first time to describe array evolution. Based on the combination of cluster evolution and time-variant channel parameters, some vital statistical properties are derived and analyzed, including space–time correlation function (ST-CF), angular power spectrum density (PSD), Doppler PSD, Doppler spread (DS), frequency correlation function (FCF), and delay spread (RS). The non-stationarity in the time, space, and frequency domain is captured. It demonstrates that the airspeed, density of scatterers within clusters, and carrier frequency have an impact on statistical properties. Furthermore, twin clusters have more flexible spatial characteristics with lower power than single clusters. These conclusions can provide assistance and reference for the design and deployment of 6G UAV communication systems.

The quality of unmanned surface vehicle (USV) local path planning directly affects its safety and autonomy performance. The USV local path planning might easily be trapped into local optima. The swarm intelligence optimization algorithm is a novel and effective method to solve the path-planning problem. Aiming to address this problem, a hybrid bacterial foraging optimization algorithm with a simulated annealing mechanism is proposed. The proposed algorithm preserves a three-layer nested structure, and a simulated annealing mechanism is incorporated into the outermost nested dispersal operator. The proposed algorithm can effectively escape the local optima. Convention on the International Regulations for Preventing Collisions at Sea (COLREGs) rules and dynamic obstacles are considered as the constraints for the proposed algorithm to design different obstacle avoidance strategies for USVs. The coastal port is selected as the working environment of the USV in the visual test platform. The experimental results show the USV can successfully avoid the various obstacles in the coastal port, and efficiently plan collision-free paths.

Accurate vehicular positioning is important for intelligent and connected vehicles (ICVs). However, in urban canyons, vehicles that rely solely on global navigation satellite system (GNSS) are susceptible to factors such as signal blocking and multi-path, reducing the positioning performance. In this paper, an application-oriented cooperative map matching (CMM) method is proposed, and a low-cost Global Positioning System (GPS)/BeiDou navigation satellite system (BDS) integrated positioning system is designed. The road constraints of a real traffic environment, which simplifies the computational complexity and facilitates practical applications, are modeled. The positioning system is designed to collect and store the positioning data for experimental analysis. Static and dynamic experiments are conducted to verify the effectiveness of the CMM method. From the experimental results, the mean absolute error (MAE) and root mean square error (RMSE) of the positioning with CMM correction in the static experiment are reduced by 9.0% and 4.9%, respectively. In the dynamic experiment, compared with the original positioning error, the MAE is reduced by 44.2% while the RMSE is reduced by 24.3%. The results show that the proposed method can improve vehicular positioning accuracy effectively in both static and dynamic environments.

Vehicle taillight intention detection is an important application for perception and decision making by intelligent vehicles. However, effectively improving detection precision with sufficient real-time performance is a critical issue in practical applications. In this study, a vision-based improved lightweight approach focusing on small object detection with a multi-scale strategy is proposed to achieve application-oriented real-time vehicle taillight intention detection. The proposed real-time detection model is designed based on YOLOv4-tiny, and a spatial pyramid pooling fast (SPPF) module is employed to enrich the output layer features. An additional detection scale is added to expand the receptive field corresponding to small objects. Meanwhile, a path aggregation network (PANet) is used to improve the feature resolution of small objects by constructing a feature pyramid with connections between feature layers. An expanded dataset based on the BDD100K dataset is established to verify the performance of the proposed method. Experimental results on the expanded dataset reveal that the proposed method can increase the average precision (AP) of vehicle, brake, left-turn, and right-turn signals by 1.81, 15.16, 40.04, and 41.53%, respectively. The mean average precision (mAP) can be improved by 24.63% (from 62.20% to 86.83%) at over 70 frames per second (FPS), proving that the proposed method can effectively improve detection precision with good real-time performance.