Learning Correspondence Structures for Person Re-identification

ABSTRACT:

This paper addresses the problem of handling spatial misalignments due to camera-view changes or humanpose variations in person re-identification. We first introduce a boosting-based approach to learn a correspondence structure which indicates the patch-wise matching probabilities between images from a target camera pair. The learned correspondence structure can not only capture the spatial correspondence pattern between cameras but also handle the viewpoint or humanpose variation in individual images. We further introduce a global constraint-based matching process. It integrates a global matching constraint over the learned correspondence structure to exclude cross-view misalignments during the image patch matching process, hence achieving a more reliable matching score between images. Finally, we also extend our approach by introducing a multi-structure scheme, which learns a set of local correspondence structures to capture the spatial correspondence sub-patterns between a camera pair, so as to handle the spatial misalignments between individual images in a more precise way. Experimental results on various datasets demonstrate the effectiveness of our approach.

PROJECT OUTPUT VIDEO:

EXISTING SYSTEM:

• Most existing works focus on handling the overall appearance variations between images, while the spatial misalignment among images’ local patches is not addressed.

• Although some patch-based methods address the spatial misalignment problem by decomposing images into patches and performing an online patch-level matching.

DISADVANTAGES OF EXISTING SYSTEM:

• One major challenge for person Re-ID is the uncontrolled spatial misalignment between images due to camera-view changes or human-pose variations.

• Online patch-level matching performances are often restrained by the online matching process which is easily affected by the mismatched patches due to similar appearance or occlusion.

PROPOSED SYSTEM:

In this paper, we propose a novel framework for addressing the problem of cross-view spatial misalignments in person Re-ID. Our contributions to Re-ID are four folds.

• We introduce a correspondence structure to encode cross-view correspondence pattern between cameras, and develop a global constraint-based matching process by combining a global constraint with the correspondence structure to exclude spatial misalignments between images. These two components establish a novel framework for addressing the Re-ID problem.

• Under this framework, we propose a boosting-based approach to learn a suitable correspondence structure between a camera pair. The learned correspondence structure can not only capture the spatial correspondence pattern between cameras but also handle the viewpoint or human-pose variation in individual images.

• We further extend our approach by introducing a multi structure scheme, which first learns a set of local correspondence structures to capture the spatial correspondence sub-patterns between a camera pair, and then adaptively selects suitable correspondence structures to handle the spatial misalignments between individual images. Thus, image-wise spatial misalignments can be modeled and excluded in a more precise way.

• We release a new and challenging benchmark ROAD DATASET for person Re-ID which includes large variation of human pose and camera angle.

ADVANTAGES OF PROPOSED SYSTEM:

• Our approach has obviously improved results than the non-structure method. This indicates that proper correspondence structures can effectively improve Re-ID performances by reducing patch-wise misalignments.

• The simple-average method has similar performance to the no-structure method. This implies that an inappropriate correspondence structure reduces the Re-ID performance.

• The AC+global method can be viewed as an extended version of the adjacency-constrained search method plus a global constraint. Compared with the AC+global method, our approach still achieves obviously better performance. This indicates that our correspondence structure has stronger capability in handling spatial misalignments than the adjacency-constrained search methods.

• The non-global method has improved Re-ID performances than the non-structure method. This further demonstrates the effectiveness of the correspondence structure learned by our approach. Meanwhile, our approach also has superior performance than the nonglobal method. This demonstrates the usefulness of introducing global constraint in patch matching process.

• The improvement of our approach is coherent on different distance metrics (KISSME, kLFDA, and KMFA-R_2). This indicates the robustness of our approach in handling spatial misalignments. In practice, our proposed correspondence structure can also be combined with other features or distance metrics to further improve Re-ID performances.

SYSTEM REQUIREMENTS:

HARDWARE REQUIREMENTS:

• System           : Pentium Dual Core.

• Hard Disk      : 120 GB.

• Monitor          : 15’’ LED

• Input Devices : Keyboard, Mouse

• Ram                : 1GB.

SOFTWARE REQUIREMENTS:

• Operating system : Windows 7.

• Coding Language : MATLAB

• Tool                       : MATLAB R2013A

REFERENCE:

Weiyao Lin, Yang Shen, Junchi Yan, Mingliang Xu, Jianxin Wu, Jingdong Wang, and Ke Lu, “Learning Correspondence Structures for Person Re-identification”, IEEE Transactions on Image Processing, 2017.