Crowd Density Estimation With Convolutional Neural Networks
Event details
Date | 11.06.2018 |
Hour | 09:00 › 11:00 |
Speaker | Weizhe Liu |
Location | |
Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Bertrand Merminod
Thesis advisor: Prof. Pascal Fua
Co-examiner: Dr. Ronan Boulic
Abstract
Abstract—Crowd analysis has attracted significant attention
in past years due to its increasing application in video surveillance,
traffic control and emergency management. Like any
other problems in computer vision, crowd analysis comes with
many challenges such as occlusions, high clutter, non-uniform
distribution of people and time-consuming annotating dataset.
Scale and perspective make the problem extremely difficult to
solve. Our research goal is to address these challenges especially
perspective distortion.
In this proposal, we discuss three existing work and how they
relate to our research. We first explain how researchers deal
with this problem without using convolutional neural networks.
Then we introduce a survey of recent CNN-based methods
in addressing above mentioned problems. Finally we discuss
an example of enforcing temporal consistency into video-based
crowd counting and compare these methods with our own ideas.
Background papers
Segmentation and tracking of multiple humans in crowded environments, by Zhao, Tao, Ram Nevatia, and Bo Wu IEEE transactions on pattern analysis and machine intelligence 30.7 (2008): 1198-1211.
A survey of recent advances in cnn-based single image crowd counting and density estimation, by Sindagi, Vishwanath A., and Vishal M. Patel. Pattern Recognition Letters (2017).
Spatiotemporal modeling for crowd counting in videos by Xiong, Feng, Xingjian Shi, and Dit-Yan Yeung. 2017 IEEE International Conference on Computer Vision (ICCV). IEEE, 2017.
Exam president: Prof. Bertrand Merminod
Thesis advisor: Prof. Pascal Fua
Co-examiner: Dr. Ronan Boulic
Abstract
Abstract—Crowd analysis has attracted significant attention
in past years due to its increasing application in video surveillance,
traffic control and emergency management. Like any
other problems in computer vision, crowd analysis comes with
many challenges such as occlusions, high clutter, non-uniform
distribution of people and time-consuming annotating dataset.
Scale and perspective make the problem extremely difficult to
solve. Our research goal is to address these challenges especially
perspective distortion.
In this proposal, we discuss three existing work and how they
relate to our research. We first explain how researchers deal
with this problem without using convolutional neural networks.
Then we introduce a survey of recent CNN-based methods
in addressing above mentioned problems. Finally we discuss
an example of enforcing temporal consistency into video-based
crowd counting and compare these methods with our own ideas.
Background papers
Segmentation and tracking of multiple humans in crowded environments, by Zhao, Tao, Ram Nevatia, and Bo Wu IEEE transactions on pattern analysis and machine intelligence 30.7 (2008): 1198-1211.
A survey of recent advances in cnn-based single image crowd counting and density estimation, by Sindagi, Vishwanath A., and Vishal M. Patel. Pattern Recognition Letters (2017).
Spatiotemporal modeling for crowd counting in videos by Xiong, Feng, Xingjian Shi, and Dit-Yan Yeung. 2017 IEEE International Conference on Computer Vision (ICCV). IEEE, 2017.
Practical information
- General public
- Free
Contact
- EDIC - [email protected]