Abstract We present a Visual Place Recognition (VPR) pipeline that achieves substantially improved precision as compared with approaches commonly appearing in the literature. It is based on a standard image retrieval configuration, with an initial stage that retrieves the closest candidates to a query from a database and a second stage where the list of candidates is re-ranked. The latter is realized by the introduction of a novel geometric verification procedure that uses the activations of a pre-trained convolutional neural network. It is both remarkably simple and robust to viewpoint and condition changes. As a stand-alone, general spatial matching methodology, it could be easily added and used to enhance existing VPR approaches whose output is a ranked list of candidates. The proposed two-stage pipeline is also improved through extensive optimization of hyperparameters and by the implementation of a frame-based temporal filter that takes into account past recognition results.

中文翻译：

---

通过高级 CNN 特征的空间匹配进行视觉位置识别

摘要 我们提出了一种视觉位置识别 (VPR) 管道，与文献中常见的方法相比，它实现了显着提高的精度。它基于标准图像检索配置，初始阶段从数据库中检索与查询最接近的候选者，第二阶段对候选者列表进行重新排序。后者是通过引入一种新的几何验证程序来实现的，该程序使用预训练的卷积神经网络的激活。对于视点和条件的变化来说，它既非常简单又健壮。作为一种独立的通用空间匹配方法，它可以轻松添加并用于增强现有的 VPR 方法，其输出是候选者的排名列表。