轨迹聚类的车辆再识别方法

车辆再识别旨在从多个视域不重叠的监控视频图像中检索出身份一致的车辆.由于实际场景的光照、视角以及背景的复杂变化,车辆再识别一直是计算机视觉领域极具挑战的热点问题.针对车辆再识别任务,提出一个同时利用车辆图像对之间的视觉相似度和时空约束的双流模型(tcReID),其视觉分支使用ResNet50作为骨干网络来评估视觉相似度...

Clustered tracks enabled vehicle re-identification

Vehicle re-identification (Re-Id) aims to retrieve identical vehicle images with the query one, among those collected from non-overlapping cameras. The complex diversities in illumination, viewpoint, and background of real scenes make this task a challenging topic in computer vision research. In this paper, we present a track clustering aided vehicle Re-Id model (tcReID) that leverages both the spatial-temporal constraints and visual similarities between image pairs. The visual stream consists of a ResNet50 backbone, assessing the visual similarity, while the track stream retrieves vehicle images among clustered trajectories. In addition, a joint metric is introduced to integrate two kinds of heterogenous scores into a unified framework. Tests show that tcReID model achieves 92.82% in mAP on VeRi-776 dataset, outperforming the previous state-of-the-art methods.