结合双流网络和金字塔映射的步态识别

目前，基于深度学习的步态识别方法虽然取得了一定的进展，但数据采集和步态外观的变化仍然是实现精确步态识别所面临的挑战。为了提高网络对时空步态信息的捕捉能力，提出了一种基于步态轮廓流和步态特征差分流的双流网络结构。步态轮廓流以步态轮廓图作为输入，用来提取步态序列中包含的空间步态信息；步态特征差分流则是以步态特征差分图作为输入，用来捕获相邻步态图之间的动态信息。同时，为了充分利用步态序列中的全局和局部信息，提出了多尺度金字塔映射（multi-scale pyramid mapping，MPM）模块，并插入到各单流网络中以增强网络对全局和局部步态信息的提取能力。所提方法在步态数据集CASIA-B和OU-MVLP上的平均识别精度分别达到了87.0%和85.5%，这表明双流网络架构和MPM模块可以有效地捕获步态序列中的时空步态信息。

Gait recognition combined with two-stream network and pyramid mapping

At present, although gait recognition methods based on deep learning have made some progress, data collection and changes in gait appearance are still challenges to achieving accurate gait recognition. In order to improve the network''s ability to capture temporal and spatial gait information, this paper proposed a two-stream network architecture based on gait contour flow and gait feature differential flow. The gait contour stream toke the gait contour map as input to extract the spatial gait information contained in the gait contour sequence, and the gait feature difference stream used the gait feature difference map as input to capture the dynamic information between adjacent gait frames. At the same time, to make full use of the global and local information in the gait sequence, this paper proposed a multi-scale pyramid mapping(MPM) module and inserted it into each single-stream network to enhance the network''s ability to extract global and local gait information. The average recognition accuracy of this method on the gait datasets CASIA-B and OU-MVLP reaches 87.0% and 85.5%. Respectively, the results show that the two-stream network architecture and MPM module can effectively capture the spatiotemporal gait information in the gait sequence.