Article
Economics
Mounia Zaouche, Nikolai W. F. Bode
Summary: This study uses Bayesian hierarchical spatio-temporal models to analyze pedestrian traffic at a fine resolution. The models incorporate publicly available data, street network properties, and urban environment features. The results highlight the importance of structured space-time and time-time interaction terms, as well as the relevance of built environment features. The Bayesian framework allows for tracking changes in traffic dynamics over time, and the models can be extended for further applications.
JOURNAL OF TRANSPORT GEOGRAPHY
(2023)
Article
Computer Science, Artificial Intelligence
Xingchen Zhang, Panagiotis Angeloudis, Yiannis Demiris
Summary: Pedestrian trajectory prediction is a crucial field in computer vision, applied in various areas such as autonomous driving, robot path planning, and surveillance systems. The main technique used is pattern recognition, with the challenge of modeling social interactions and pedestrian view area. Existing studies often require additional detectors and manual annotations to handle view area and group interactions. This paper proposes a dual-branch spatio-temporal graph neural network that automatically models view area and group interactions. Experimental results show competitive performance in predicting socially acceptable trajectories.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Yunpeng Chang, Zhigang Tu, Wei Xie, Bin Luo, Shifu Zhang, Haigang Sui, Junsong Yuan
Summary: This study tackles anomaly detection in videos by exploring a novel convolution autoencoder architecture that separates spatio-temporal representations to capture abnormal events. By simulating differences in appearance and motion behaviors, an effective anomaly detection method is proposed, achieving state-of-the-art performance on multiple public datasets.
PATTERN RECOGNITION
(2022)
Article
Engineering, Civil
Kinjal Dasgupta, Arindam Das, Sudip Das, Ujjwal Bhattacharya, Senthil Yogamani
Summary: This paper proposes a fusion model for pedestrian detection using RGB and thermal images, which improves the performance of pedestrian detection through multimodal fusion and deep network architecture. Experimental results show that the proposed model achieves better results than existing methods on multiple datasets.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Sarfraz Ahmed, Ammar Al Bazi, Chitta Saha, Sujan Rajbhandari, M. Nazmul Huda
Summary: With the increase in the use of Autonomous Vehicles on public roads, it is necessary for these vehicles to operate safely. While pedestrian detection has advanced in accuracy, pedestrian intent prediction still requires further research to reach human-level performance. This study presents a novel approach for multi-scale pedestrian intent prediction using 2D pose estimation and LSTM architecture. Experimental results on popular datasets show that the proposed technique outperforms state-of-the-art methods, achieving up to 94% accuracy while maintaining a comparable run-time.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Francisco Barranco, Cornelia Fermueller, Yiannis Aloimonos, Eduardo Ros
Summary: This paper investigates the impact of avoiding optical flow estimation on structure recovery, and proposes a new method based on image gradients to solve 3D motion problems by reformulating the positive-depth constraint. Experimental results show that the method achieves higher accuracy and outperforms existing techniques based on normal flow for 3D motion estimation.
PATTERN RECOGNITION
(2021)
Article
Computer Science, Artificial Intelligence
Dongfang Yang, Haolin Zhang, Ekim Yurtsever, Keith A. Redmill, Umit Ozguner
Summary: This study presents a novel neural network architecture to predict pedestrian crossing intention by fusing different spatio-temporal features. By optimally combining various phenomena such as RGB imagery sequences, semantic segmentation masks, and ego-vehicle speed, the proposed method achieves state-of-the-art performance.
IEEE TRANSACTIONS ON INTELLIGENT VEHICLES
(2022)
Article
Engineering, Electrical & Electronic
Rodney V. Fonseca, Rogerio G. Negri, Aluisio Pinheiro, Abdourrahmane Mahamane Atto
Summary: In this article, we introduce the wavelet energies correlation screening (WECS), an unsupervised method to detect spatio-temporal changes on multitemporal SAR images. The procedure is based on wavelet approximation for the multitemporal images, wavelet energy apportionment, and ultrahigh-dimensional correlation screening for the wavelet coefficients. WECS's performance is demonstrated on simulated multitemporal image data and evaluated on a time series of 85 Sentinel-1 images of a forest region at the border of Brazil and French Guiana. Comparisons with existing change detection methods highlight the superiority of the proposed method in terms of change detection accuracy. Additionally, the introduced method has a simple architecture and low computational cost.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Engineering, Civil
Daniel Gomez, Shirley Rietdyk, Shirley J. Dyke
Summary: Traditional modeling approaches neglect the temporal and spatial kinematic changes when pedestrians walk on a vibrating surface, compromising comfort in service conditions. This study demonstrates an interaction between gait and bridge kinematics, suggesting further research is needed to understand this relationship more comprehensively.
Article
Computer Science, Artificial Intelligence
Brais Bosquet, Manuel Mucientes, Victor M. Brea
Summary: Object detection using convolutional neural networks has achieved unprecedented levels of accuracy, but there is still room for improvement in detecting small objects. Utilizing spatial information alongside temporal video data is a new trend that can potentially enhance overall object detection performance. STDnet-ST is an end-to-end spatio-temporal convolutional neural network designed for detecting small objects in video, achieving state-of-the-art results on various video datasets.
PATTERN RECOGNITION
(2021)
Article
Chemistry, Analytical
Hesham Alghodhaifi, Sridhar Lakshmanan
Summary: Ensuring the safety of intelligent vehicles in interactions with pedestrians is challenging due to pedestrians' unpredictable movements. This paper proposes a novel graph-based trajectory prediction model called HSTGA to address the limitations of existing studies. HSTGA extracts spatial features of vehicle-pedestrian interactions and combines them with LSTM and graph attention networks to predict trajectories more accurately.
Article
Mathematics, Interdisciplinary Applications
Hongzhi Zhou, Gan Yu
Summary: This paper proposes a fast pedestrian detection algorithm by combining autoencoding neural network and AdaBoost, and introduces a two-input AdaBoost-DBN classification algorithm to address the issues of low accuracy and high missed detection rate. Additionally, motion compensation and frame reconstruction are used to solve the problem of video fusion playback not being smooth.
Article
Cell Biology
Paolo Armando Gagliardi, Benjamin Graedel, Marc-Antoine Jacques, Lucien Hinderling, Pascal Ender, Andrew R. Cohen, Gerald Kastberger, Olivier Pertz, Maciej Dobrzynski
Summary: This article introduces a computational method called ARCOS for the detection and quantification of collective signaling in cell collectives. The method is demonstrated on cell and organism collectives with space-time correlations of varying scales. The study reveals the importance of intercellular signaling for the functionality and adaptation of cell collectives to environmental challenges.
JOURNAL OF CELL BIOLOGY
(2023)
Article
Computer Science, Information Systems
Bogdan Ilie Sighencea, Ion Rares Stanciu, Catalin Daniel Caleanu
Summary: Predicting pedestrian trajectories in urban scenarios is a challenging task with various applications. This paper proposes an attention-based spatio-temporal graph neural network that includes a spatial graph neural network (SGNN) and a temporal graph neural network (TGNN) for trajectory prediction. The proposed approach outperforms the social-STGCNN model in terms of accuracy and efficiency.
Article
Environmental Sciences
Manhong Tu, Weixing Zhang, Jingna Bai, Di Wu, Hong Liang, Yidong Lou
Summary: Processing GPS data from 700 stations in China during Typhoon Lekima revealed variations in PWV and the performance of different weather models. PWV increased as the typhoon approached and decreased as it left; radiosonde, GFS, and ERA5 tended to overestimate PWV compared to GPS. Analysis of PWV during the typhoon event showed different variability patterns.