Method Study of Cross-view Activity Recognition Based on Transfer Learning
Abstract: Activity recognition is one of the most important problems in the fields of computer vision and pattern recognition, has great potential to be applied in real applications like intelligent visual surveillance. The cross-view problem is the key in the topic of activity recognition, becoming the bottleneck to further booming the performance of activity recognition. We focus on cross-view activity recognition. We hope to collaborate with each other and try to find new strategies and algorithms to achieve a breakthrough in this topic. Specifically, we would try to find automatic features for hierarchical representation, multi-cue fusion for transfer learning. We hope to find not only solutions in the topic of cross-view activity recognition, but also solutions to cross-view video analysis.
Keywords: transfer learning; deep learning; activity analysis
Contact:
ZHANG Zhaoxiang
E-mail: zhaoxiang.zhang@ia.ac.cn