Feature Representation for Crowd Counting by Regression: A Review

Abstract: Crowd density estimation is crucial for intelligent video surveillance to help in control and management of crowds for safety. Crowd density analysis is related to the crowd feature extraction. This paper presents a review on feature representation for crowd counting by regression to construct intermediate input to a regression model. The comparison between feature extraction techniques with experimental results shows that statistical methods are easy to apply and the extracted features are stable to the circumstance change. With the great accuracy the method is efficient for the crowd density estimation.

Keywords: Performance Evaluation of Tracking and Surveillance (PETS) Dataset, blob count, Fractal dimension, Gray level Co-occurrence Matrix (GLCM).