A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Monitoring Pedestrian Flow on Campus with Multiple Cameras using Computer Vision and Deep Learning Techniques
https://orcid.org/http://orcid.org/0000-0003-1758-675X
https://orcid.org/http://orcid.org/0000-0002-1722-2617
This paper proposes a robust method for multi-camera person re-identification (ReID), which can accurately track pedestrians across non-overlapping cameras. Closed-circuit television (CCTV) are widely used to capture pedestrian movement in different places. By integrating CCTV with computer vision and deep learning techniques, trajectory of individual pedestrian can be efficiently acquired for analyzing pedestrian walking behaviors. Many existing ReID methods aim to extract discriminative human features to distinguish a person from others. Recent state-of-the-art performance is achieved mostly by obtaining fine features from each body part. However, these part-based feature extraction methods did not consider which parts are more useful for person ReID. Therefore, this paper proposes a weighted-parts feature extraction method, such that features of specific body parts are more influential to identity prediction. After comparing the performances of utilizing each part alone, several parts are considered more view-invariant and discriminative. Higher weights are then imposed on these specific parts to extract more useful human features for person ReID. Experimental results with videos on a college campus show that the ReID accuracy of our proposed method notably outperforms many existing ones.
Monitoring Pedestrian Flow on Campus with Multiple Cameras using Computer Vision and Deep Learning Techniques
https://orcid.org/http://orcid.org/0000-0003-1758-675X
https://orcid.org/http://orcid.org/0000-0002-1722-2617
This paper proposes a robust method for multi-camera person re-identification (ReID), which can accurately track pedestrians across non-overlapping cameras. Closed-circuit television (CCTV) are widely used to capture pedestrian movement in different places. By integrating CCTV with computer vision and deep learning techniques, trajectory of individual pedestrian can be efficiently acquired for analyzing pedestrian walking behaviors. Many existing ReID methods aim to extract discriminative human features to distinguish a person from others. Recent state-of-the-art performance is achieved mostly by obtaining fine features from each body part. However, these part-based feature extraction methods did not consider which parts are more useful for person ReID. Therefore, this paper proposes a weighted-parts feature extraction method, such that features of specific body parts are more influential to identity prediction. After comparing the performances of utilizing each part alone, several parts are considered more view-invariant and discriminative. Higher weights are then imposed on these specific parts to extract more useful human features for person ReID. Experimental results with videos on a college campus show that the ReID accuracy of our proposed method notably outperforms many existing ones.
Monitoring Pedestrian Flow on Campus with Multiple Cameras using Computer Vision and Deep Learning Techniques
https://orcid.org/http://orcid.org/0000-0003-1758-675X
https://orcid.org/http://orcid.org/0000-0002-1722-2617
This paper proposes a robust method for multi-camera person re-identification (ReID), which can accurately track pedestrians across non-overlapping cameras. Closed-circuit television (CCTV) are widely used to capture pedestrian movement in different places. By integrating CCTV with computer vision and deep learning techniques, trajectory of individual pedestrian can be efficiently acquired for analyzing pedestrian walking behaviors. Many existing ReID methods aim to extract discriminative human features to distinguish a person from others. Recent state-of-the-art performance is achieved mostly by obtaining fine features from each body part. However, these part-based feature extraction methods did not consider which parts are more useful for person ReID. Therefore, this paper proposes a weighted-parts feature extraction method, such that features of specific body parts are more influential to identity prediction. After comparing the performances of utilizing each part alone, several parts are considered more view-invariant and discriminative. Higher weights are then imposed on these specific parts to extract more useful human features for person ReID. Experimental results with videos on a college campus show that the ReID accuracy of our proposed method notably outperforms many existing ones.
Monitoring Pedestrian Flow on Campus with Multiple Cameras using Computer Vision and Deep Learning Techniques
Lecture Notes in Civil Engineering
Ha-Minh, Cuong (editor) / Dao, Dong Van (editor) / Benboudjema, Farid (editor) / Derrible, Sybil (editor) / Huynh, Dat Vu Khoa (editor) / Tang, Anh Minh (editor) / Wong, Peter Kok-Yiu (author) / Cheng, Jack C. P. (author)
2019-10-11
6 pages
Article/Chapter (Book)
Electronic Resource
English
Computer vision , Deep learning , Human feature extraction , Multi-camera re-identification , Pedestrian flow analytics Engineering , Geoengineering, Foundations, Hydraulics , Sustainable Development , Landscape/Regional and Urban Planning , Structural Materials , Building Construction and Design , Transportation Technology and Traffic Engineering
Computer Vision and Multi-Object Tracking for Traffic Measurement from Campus Monitoring Cameras
| British Library Conference Proceedings | 2021
Pedestrian Flows Characterization and Estimation with Computer Vision Techniques
| DOAJ | 2023
Automated Collection of Pedestrian Data Through Computer Vision Techniques
| British Library Online Contents | 2012