Person Re-Identification across Non-Overlapping Cameras Based on Two-Stage Framework
Feigang Tan, Xiaoju Zhao, Kaiyuan Liu, Quanmi Liao · 2018
The appearance characteristics of persons provide effective discrimination information for person re-identification across non-overlapping cameras. However, there are many factors that make person appearances differences with strong uncertainty in the actual monitoring scene. Such as shooting angle and illumination changes. The statistics show that some persons are easier to distinguish among non-overlapping cameras, while others are more difficult to distinguish. To reduce the persons interference which is easy to distinguish, and increase the accuracy of the algorithm. We propose a person re-identification across non-overlapping cameras based on two-stage framework. Firstly, the simple feature is used to quickly eliminate the easily distinguishable persons. Then the prototype similarity feature is used to re-identification the remaining persons. Experiments on public datasets show that the proposed algorithm outperforms other methods.