Detecting social groups in crowds by means of a Correlation Clustering procedure on people trajectories. According to my observation, "split" "together" "meet" are three meaningful features to describe interaction of two trajectories.
3. Social Groups Detection
• Detecting social groups in crowds by means of
a Correlation Clustering procedure on people
trajectories
Solera, Francesco, Simone Calderara, and Rita Cucchiara. "Socially Constrained Structural
Learning for Groups Detection in Crowd." (2015).
5. Identification of social groups
• Two pedestrians are considered similar if their
corresponding motion parameter values
satisfy the following criteria
-Distance between pedestrians < 2m
-Relative speed difference < 0.4m/s
-Relative difference in direction of travel < 3 deg
McPhail, Clark, and Ronald T. Wohlstein. "Using film to analyze pedestrian
behavior." Sociological Methods & Research 10.3 (1982): 347-375.
8. Space Issue
• False negative
– (15, 16)(17, 18)(73, 74) Exhibit more dilated
patterns due to groups in open spaces
• False positive
– (36, 37)(64, 65)(66,67) Groups in crowded scenes
Distance should be re-weight
9. Density Invariance
• Groups in very crowded scenes will be more
closed and compact, while groups in open
spaces will tend to exhibit more dilated
patterns
M. Moussaid, N. Perozo, S. Garnier, D. Helbing, and G. Theraulaz, “The walking behaviour of
pedestrian social groups and its impact on crowd dynamics,” PLoS ONE, vol. 5, Apr. 2010.
10. Clustering
• Re-weight, from space to density
• Cost-efficient, do not need to consider members out of
groups
μ
μ
μ
σ
σ
σ
11. Framework
Clustering
Feature Extraction
Grouping people walking close to each other with similar
velocities
Distance+ Degree
Social-psychology theory
-Distance between pedestrians < 2m
-Relative speed difference < 0.4m/s
-Relative difference in direction of travel < 3 deg
Ge, Weina, Robert T. Collins, and R. Barry Ruback. "Vision-based analysis of small groups in pedestrian
crowds." Pattern Analysis and Machine Intelligence, IEEE Transactions on 34.5 (2012): 1003-1016.
13. Degree Feature
• According to my observation, "split"
"together" "meet" are three meaningful
features to describe interaction of two
trajectories.
14. Degree Feature
Split Meet Together Together
․Adaptive sub-sequence?
․Feature Dimension?
․Definition of Split, Meet
and together?
15. Two Features
• Not only orientation, spatial behaviors are
created by people who sustain the shared
interaction space between them
• Use "degree" and "physical distance" to
represent interaction between trajectories
Zhang, Yanhao, et al. "Formation Period Matters: Towards Socially Consistent Group Detection via
Dense Subgraph Seeking." Proceedings of the 5th ACM on International Conference on Multimedia
Retrieval. ACM, 2015.
16. Experiment
File Feature Precision Recall
1manko3 4 Features 72.96 % 70.52 %
2 Features 72.2 % 70.36 %
1manko29 4 Features 84 % 91.3 %
2 Features 88 % 88 %
2jiansha5 4 Features 75.47 % 76.92 %
2 Features 69.81 % 72.55 %