常见用于Place Recognition的数据库的总结

1. SPED_900

- 特点： 超过30,000个监控摄像头，遍及全世界，从2006年开始到现在（还会继续持续），
- GT： 这个主要用于训练网络的，先不管
- 用途： 主要用于训练
- 获取： 找作者要的
提出的论文：

Deep Learning Features at Scale for Visual Place Recognition

2. bovisa

3. UA Compus

4. St Lucia

- 特点： 同一条路线，一天中的不同时间段（20-25 minutes starting at 8:45, 10:00, 12:10, 14:10, 15:45）
- GT： 数据库给了GPSlog文件
- 用途： 主要用于验证光照的影响，我的实验主要打算用于测试
- 获取： 我有啊
contains 10 subdatasets with images captured from the same route which is traversed at different times of the day from morning to afternoon.

剔除的论文：

A. Glover, W. Maddern, M. Milford, and G. Wyeth, “FAB-MAP + RatSLAM : Appearance-Based SLAM for Multiple Times of Day,” in Proc. of IEEE Intl. Conf. on Robotics and Automation (ICRA), 2010

The St. Lucia dataset [31] has been recorded from inside a car moving through a suburb in Brisbane at 5 different times during a day, and also on different days over a time of two weeks. It features mild viewpoint variations due to slight changes in the exact route taken by the car. More significant appearance changes due to the different times of day can be observed, as well as some changes due to dynamic objects such as traffic or cars parked on the street

5. NordLand

- 特点：Train旅行中拍摄的，有一年四季，视频长10小时左右。
- GT：暂时不知道
- 用途：验证appearance changes，常用于测试
- 获取：我有啊

提出该数据库的论文：

N. S¨ underhauf, P. Neubert, and P. Protzel. Are we there yet? challenging seqslam on a 3000 km journey across all four seasons. Proc. of Workshop on Long-Term Autonomy, IEEE International Conference on Robotics and Automation (ICRA), page 2013, 2013.

The Nordland dataset, extracted from the TV documentary “Nordlandsbanen - Minutt for Minutt” produced by the Norwegian Broadcasting Corporation NRK consists of a 728 km long train journey connecting the cities of Trondheim and Bodø in Norway. The sequence was recorded once in each season, and hence it contains challenging appearance changes, Additionally, it provides different weather conditions due to the large length of the dataset (the sequences are 10 hour long approximately). We generate triplets by providing two images from the same place in different seasons, and an image from another location in any season (we check that frames are actually from different places using the included GPS ground truth).

这里提到的三元组的训练方式出自：

Training a Convolutional Neural Network for Appearance-Invariant Place Recognition

6. New College

7. City Center