A Spatiotemporal Probabilistic Graphical Model Based on Adaptive Expectation-Maximization Attention for Individual Trajectory Reconstruction Considering Incomplete Observations.

Spatiotemporal information on individual trajectories in urban rail transit is important for operational strategy adjustment, personalized recommendation, and emergency command decision-making. However, due to the lack of journey observations, it is difficult to accurately infer unknown information from trajectories based only on AFC and AVL data. To address the problem, this paper proposes a spatiotemporal probabilistic graphical model based on adaptive expectation maximization attention (STPGM-AEMA) to achieve the reconstruction of individual trajectories. The approach consists of three steps: first, the potential train alternative set and the egress time alternative set of individuals are obtained through data mining and combinatorial enumeration. Then, global and local potential variables are introduced to construct a spatiotemporal probabilistic graphical model, provide the inference process for unknown events, and state information about individual trajectories. Further, considering the effect of missing data, an attention mechanism-enhanced expectation-maximization algorithm is proposed to achieve maximum likelihood estimation of individual trajectories. Finally, typical datasets of origin-destination pairs and actual individual trajectory tracking data are used to validate the effectiveness of the proposed method. The results show that the STPGM-AEMA method is more than 95% accurate in recovering missing information in the observed data, which is at least 15% more accurate than the traditional methods (i.e., PTAM-MLE and MPTAM-EM).

A Survey on Vision Transformer.

Transformer, first applied to the field of natural language processing, is a type of deep neural network mainly based on the self-attention mechanism. Thanks to its strong representation capabilities, researchers are looking at ways to apply transformer to computer vision tasks. In a variety of visual benchmarks, transformer-based models perform similar to or better than other types of networks such as convolutional and recurrent neural networks. Given its high performance and less need for vision-specific inductive bias, transformer is receiving more and more attention from the computer vision community. In this paper, we review these vision transformer models by categorizing them in different tasks and analyzing their advantages and disadvantages. The main categories we explore include the backbone network, high/mid-level vision, low-level vision, and video processing. We also include efficient transformer methods for pushing transformer into real device-based applications. Furthermore, we also take a brief look at the self-attention mechanism in computer vision, as it is the base component in transformer. Toward the end of this paper, we discuss the challenges and provide several further research directions for vision transformers.

Face detection for rail transit passengers based on single shot detector and active learning.

COVID-19 spreads rapidly among people, so that more and more people are wearing masks in rail transit stations. However, the current face detection algorithms cannot distinguish between a face wearing a mask and a face not wearing a mask. This paper proposes a face detection algorithm based on single shot detector and active learning in rail transit surveillance, effectively detecting faces and faces wearing masks. Firstly, we propose a real-time face detection algorithm based on single shot detector, which improves the accuracy by optimizing backbone network, feature pyramid network, spatial attention module, and loss function. Subsequently, this paper proposes a semi-supervised active learning method to select valuable samples from video surveillance of rail transit to retrain the face detection algorithm, which improves the generalization of the algorithm in rail transit and reduces the time to label samples. Extensive experimental results demonstrate that the proposed method achieves significant performance over the state-of-the-art algorithms on rail transit dataset. The proposed algorithm has a wide range of applications in rail transit stations, including passenger flow statistics, epidemiological analysis, and reminders of passenger who do not wear masks. Simultaneously, our algorithm does not collect and store face information of passengers, which effectively protects the privacy of passengers.