基于改进PVEN模型的车辆再识别算法

Vehicle Re-Identification Algorithm Based on Improved PVEN

随着城市交通快速发展,监控场景下的车辆再识别任务成为计算机视觉领域的热点研究问题。车辆再识别是指在非重叠视角域、跨摄像机网络下识别同一身份车辆。与行人再识别相比,类间相似性和类内差异性使得车辆再识别任务面临更大的挑战。为提高车辆再识别的准确率,提出一种基于改进PVEN(parsing-based view-aware embedding network)模型的车辆再识别算法。算法整体采用了PVEN模型框架:首先,针对原模型中特征提取网络简单、难以提取车辆有效特征,提出融合注意力机制使网络能够关注到更有辨识力的特征信息;其次,优化网络的训练过程,将不同车辆图像特征之间的欧氏距离作为基础相似性度量,建立联合目标函数,并分别分析通道注意力和空间注意力两个模块对网络性能的影响。在VeRi776数据集上进行相关实验,所提算法在评价指标MAP、Rank-1的表现为80.0%和96.2%,较原模型分别提高约0.5%和0.6%,证明了所提算法的有效性。

In order to improve the vehicle recognition accuracy,this paper proposes a model based on improved PVEN vehicle re-identification algorithm,algorithm integrated with PVEN(parsing-based view-aware embedding network)model frameworkfirst of all,according to the original model in feature extraction is simple,it is difficult to extract the vehicle network effective characteristics of fusion attention mechanism enables the network to focus on the more discrimination characteristic information.Secondly,the network training process is optimized,the Euclidean distance between the joint features of the target vehicles was taken as the similarity measure,the joint objective function was established,and the effects of channel attention and spatial attention on the network performance are analyzed respectively.Experiments on the Veri776 dataset show that the performance of the proposed model is about 0.5%and 0.6%better than that of the original model on the evaluation index MAP and Rank-1,respectively,which demonstrate the effectiveness of the proposed algorithm.