基于局部特征和焦点融合的车辆重识别算法

由于城市监控中存在大量相似的车辆,造成了车辆重识别匹配率低。车头、车窗、车顶等局部特征是相似车辆细微差异性的所在。根据车辆检测算法卷积特征热力图注意力分布特性,提出了针对车辆局部特征区域检测的MCRF-SSD算法,并与GMM-EM聚类算法相结合,检测性能在公开的数据集上均优于目前主流算法。同时为了增大类间距离、缩小类内距离将arcface损失函数引入到了特征提取阶段。为了提高车辆重识别匹配性能,在全局特征与局部特征融合阶段提出了一种保留特征图空间分布的焦点融合（Ffs）方法,并引入了一个可学习参数,提高了特征融合效率。实验结果表明,所提出的算法在公开的VehicleID和VeRi数据集中性能表现优于目前性能最优的方案。

Vehicle re-identification based on local feature and focus fusion

There are many similar vehicles in city monitoring, which brings great challenges to vehicle re-identification. Local features such as front, window and roof are the subtle differences of similar vehicles. According to the attention characteristics of the thermal map of the vehicle detection algorithm, a MCRF-SSD algorithm is proposed to detect the local feature area of the vehicle, and combines it with GMM-EM clustering algorithm. The detection performance is better than the current mainstream algorithm on the open data set.At the same time, in order to increase the inter-instance and reduce the intra-instance, the Arcface loss function is introduced into the feature extraction stage. In order to improve the performance of vehicle re recognition, in the stage of global feature and local feature fusion, a focus fusion structure (FFS) method is proposed, which can preserve the spatial distribution of feature graph, and a learnable parameter is introduced to improve the efficiency of feature fusion. Experimental results show that the performance of the proposed algorithm is better than that of the current best performance scheme in public VehicleID and VeRi datasets.