联合语义分割与注意力机制的行人再识别模型

Pedestrian Re-Identification Model Combining Semantic Segmentation and Attention Mechanism

受行人姿态变化、光照视角、背景变换等因素的影响,现有行人再识别模型通常对数据集中的行人分成若干块提取图像的局部特征进行辨识以提高识别精度,但存在人体局部特征不匹配、容易丢失非人体部件的上下文线索等问题。构建一种改进的行人再识别模型,通过将人体语义解析网络的局部特征进行对齐,增强行人语义分割模型对图像中行人任意轮廓的建模能力,利用局部注意力网络捕捉非人体部分丢失的语境线索。实验结果表明,该模型在Market-1501、DukeMTMC和CUHK03数据集上的平均精度均值分别达到83.5%、80.8%和92.4%,在DukeMTMC数据集上的Rank-1为90.2%,相比基于注意力机制、行人语义解析和局部对齐网络的行人再识别模型具有更强的鲁棒性和迁移性。

Pedestrian identification results are easily affected by pedestrian posture changes,illumination perspective,background transformation and other factors.To reduce such interference,the existing pedestrian re-identification models usually divide the pedestrians in a dataset into several pieces to extract the local features of the image and improve the identification accuracy,but this also presents new problems such as the mismatch between local features of the human body and the loss of contextual clues of non-human parts.In order to solve the above problems,an improved pedestrian re-identification model is proposed.By aligning the local features of the human semantic parsing network,the semantic segmentation model can perform better in modeling arbitrary contours of pedestrians in the image.The local attention network is also used to capture the lost contextual clues of non-human body parts.The experimental results show that the proposed model displays an average accuracy of 83.5%on Market-1501,80.8%on DukeMTMC,and 92.4%on CUHK03.The Rank-1 value on the DukeMTMC dataset is 90.2%.Compared with the pedestrian re-identification models based on attention mechanism,pedestrian semantic parsing network or Partial Alignment Network(PAN),the proposed model has higher robustness and mobility.