作为分布式摄像机物联网的重要应用之一,行人重识别任务是从不同摄像头拍摄的监控录像中寻找并识别出已出现在某个摄像机视野的特定行人.为应对低分辨率、遮挡以及姿态、视点和光照变化情形下行人细粒度的重识别问题,提出了一种新的数据驱动模型来推断多摄像机的逻辑拓扑结构模型,重识别不同多摄像机捕获的行人;在模型中,用时间延迟互信息模型(time delayed mutual information,TDMI)实现多摄像机的逻辑拓扑推理,利用训练好的深度卷积神经网络(deep convolutional neural network,DCNN)来提取特征,用可命名模型发现外观属性,并基于该属性的结构输出进行分类,实现行人重识别.实验结果表明,与其他模型相比,所提模型在结构化语义属性输出和行人重识别精度方面都达到了较好的水平.%To make fine-grained person re-identification against very low resolution, occlusions and pose, viewpoint and illumination changes, a novel data-driven model is proposed to infer multi-cameras logical topology and re-identify persons captured by different cameras.In our model, a time delayed mutual information (TDMI) model is employed to address multi-cameras logical topology inference.Then, a well-trained deep convolutional neural network (DCNN) is utilized to extract features.Finally, a nameability model is employed to discover attributes and a classifier based on structural output of attributes is designed to address the person re-identification.Experimental results show that compared with other models, our model presents better performance by imposing semantic restrictions onto the generation of human specific attributes with structural output and employing deep learning model to generate features without supervision for attributes learning model.
展开▼
