Classification of moving humans using eigen-features and support vector machines

Authors:
Sijun Lu;Jian Zhang;David Feng
Affiliations:
National ICT, Australia;National ICT, Australia;School of Information Technology, The University of Sydney, Australia
Venue:
CAIP'05 Proceedings of the 11th international conference on Computer Analysis of Images and Patterns
Year:
2005

Citing 6
Cited 4

A training algorithm for optimal margin classifiers

COLT '92 Proceedings of the fifth annual workshop on Computational learning theory
A Real-Time System for Classification of Moving Objects

ICPR '02 Proceedings of the 16 th International Conference on Pattern Recognition (ICPR'02) Volume 3 - Volume 3
A short introduction to learning with kernels

Advanced lectures on machine learning
Boosting Object Detection Using Feature Selection

AVSS '03 Proceedings of the IEEE Conference on Advanced Video and Signal Based Surveillance
Detection and Recognition of Moving Objects Using Statistical Motion Detection and Fourier Descriptors

ICIAP '03 Proceedings of the 12th International Conference on Image Analysis and Processing
LIBSVM: A library for support vector machines

ACM Transactions on Intelligent Systems and Technology (TIST)

Collaborative signal processing for target tracking in distributed wireless sensor networks

Journal of Parallel and Distributed Computing
Detecting unattended packages through human activity recognition and object association

Pattern Recognition
Collaborative Target Classification for Image Recognition in Wireless Sensor Networks

ADMA '07 Proceedings of the 3rd international conference on Advanced Data Mining and Applications
Distributed visual-target-surveillance system in wireless sensor networks

IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper describes a method of categorizing the moving objects using eigen-features and support vector machines. Eigen-features, generally used in face recognition and static image classification, are applied to classify the moving objects detected from the surveillance video sequences. Through experiments on a large set of data, it has been found out that in such an application the binary image instead of the normally used grey image is the more suitable format for the feature extraction. Different SVM kernels have been compared and the RBF kernel is selected as the optimal one. A voting mechanism is employed to utilize the tracking information to further improve the classification accuracy. The resulting labeled object trajectories provide important hints for understanding human activities in the surveillance video.