Person Re-identification In Intelligent Visual Surveillance

Person Re-Identification(ReID)is one of the most vital tasks in an intelli-gent visual surveillance system,which aims at associating the same pedestrian across multiple camera views,and extending the visual surveillance from single camera to camera network.Based on the previous substantive research work,we propose to improve the RelD performance from three perspectives,includ-ing the generalization ability in metric learning on small dataset,the feature engineering and fusion model,and the feature sequence extraction and match-ing model.From the perspective of the generalization ability in metric learning on small dataset,we propose to accommodate the regularization strategy to en-hance the robustness of the methods.Metric learning plays a critical role in person re-identification problem.Unfortunately,due to the small size of train-ing data,the metric learning used in this scenario suffers from over-fitting which leads to degenerated performance.In this paper,we investigate the effect of reg-ularization in metric learning for person re-identification.Concretely we for-mulate the distance function from three perspectives and hence present four dif-ferent regularized metric learning methods.Experiments on two popular bench-mark data sets VIPeR and CUHK01 validate the effectiveness of our proposed regularization approaches.From the perspective of feature engineering and fusion model,we propose a unified framework of statistical local feature extraction and combination for ReID.Re-identifying individual across non-overlapping camera views is one of challenging problems in surveillance video analysis.The difficulties mainly come from the large appearance variations caused by camera view angle,human pose,illumination,and occlusion.Recently,extensive efforts have been cast into addressing this problem by developing invariant features or discriminative distance metrics.However,there is still a lack of systematic evaluations on the pipeline for feature extraction and combination.In this paper,we propose a spa-tial pyramid based statistical feature extraction framework as a unified pipeline of feature extraction and combination for person ReID,and systematically eval-uate the configuration details in feature extraction and the strategies in feature combination.Extensive experiments on benchmark datasets demonstrate the critical components in feature extraction.Moreover,by combining multiple features,our proposed approach can yield state-of-the-art performance.From the perspective of feature sequence extraction and matching model,we propose a novel end-to-end trainable framework for person ReID,which can jointly learn context-aware feature sequences and perform sequences compar-ison with dual attention mechanism.Typical ReID methods usually describe each pedestrian with a single feature vector and match them in a task-specific metric space.However,the methods based on a single feature vector are not suf-ficient enough to overcome visual ambiguity,which frequently occurs in real scenario.In this paper,we propose a novel end-to-end trainable framework,called Dual ATtention Matching network(DuATM),to learn context-aware fea-ture sequences and perform attentive sequence comparison simultaneously.The core component of our DuATM framework is a dual attention mechanism,in which both intra-sequence and inter-sequence attention strategies are used for feature refinement and feature-pair alignment,respectively.Thus,detailed vi-sual cues contained in the intermediate feature sequences can be automatically exploited and properly compared.We train the proposed DuATM network as a siamese network via a triplet loss assisted with a de-correlation loss and a cross entropy loss.We conduct extensive experiments on both image and video based ReID benchmark datasets.Experimental results demonstrate the significant ad-vantages of our approach compared to the state-of-the-art methods.