Toulouse Campus Surveillance Dataset: scenarios, soundtracks, synchronized videos with overlapping and disjoint views

The Toulouse Campus surveillance Dataset, named ToCaDa, contains two sets of 25 temporally synchronized videos corresponding to two scripted scenarios. With the help of about 50 persons (actors and camera holders), these videos were shot on July 17th 2017 at 9:50 a.m. and 11:04 a.m. respectively. Among the cameras: • 9 were located inside the main building and shot from the windows at different floors. All these cameras are focusing the car park and the path leading to the main entrance of the building with large overlapping fields of view. • 8 were located in front of the building and filmed it with large overlapping fields of view. • 8 cameras were arranged further, scattered around the university campus. Each of their views is disjoint from all the others. About 20 actors were asked to follow two realistic scenarios by performing scripted actions, like driving a car, walking, entering or leaving a building, or holding an item in hand while being filmed. In addition to ordinary actions, some suspicious behaviors are present. Irregularities: Due to the wide variety of devices used during the shooting of the two scenarios, issues were encountered on some cameras, leading to videos where a few seconds are lacking. To ensure temporal synchronization between videos, black frames were added on the missing intervals of time. We list these particular videos and their lacking times below: F1C3: the first 66 seconds are missing. F1C5: the first 2 seconds are missing. F1C8: the first 3 seconds are missing. F1C13: the first 10 seconds are missing. F1C15: the first second is missing. F1C19: the first second is missing. F2C1: the video is accelerated and only lasts a few seconds. We thus did not provide it. F2C6: lacks from 4:01 to 4:12 and from 4:25 to 4:28. F2C16: lack from 5:15 to 5:26. Some videos were recorded with mobile devices whose pixel resolution was lower than 1920 x 1080: F1C3 and F2C3: pixel resolution is 1280 x 720. F1C4 and F2C4: pixel resolution is 640 x 480. F1C15 and F2C15: pixel resolution is 1280 x 720. F1C20 and F2C20: pixel resolution is 1440 x 1080. More detailed information about the position of the cameras can be found on the following link: http://ubee.enseeiht.fr/dokuwiki/doku.php?id=public:tocada Citation T. Malon, G. Roman-Jimenez, P. Guyot, S. Chambon, V. Charvillat, A. Crouzil, A. Péninou, J. Pinquier, F. Sèdes and C. Sénac, Toulouse campus surveillance dataset: scenarios, soundtracks, synchronized videos with overlapping and disjoint views, ACM Multimedia Systems Conference, 2018.

图卢兹校区监控数据集（Toulouse Campus surveillance Dataset，简称ToCaDa）包含两组各25条时间同步视频（temporally synchronized videos），对应两个脚本化场景（scripted scenarios）。本次拍摄由约50名人员（演员与摄像师）协助完成，分别于2017年7月17日上午9:50与11:04进行。

相机部署情况如下：
• 9台部署于主楼内部，通过不同楼层的窗口取景，所有相机均聚焦于停车场及通往主楼正门的通道，视场（field of view）存在大面积重叠。
• 8台部署于楼宇前方，以大面积重叠的视场拍摄该建筑。
• 另有8台相机布置于更远位置，分散在大学校园内，各相机的视场彼此完全互不重叠（disjoint）。

约20名演员按照脚本执行两类写实场景动作，包括驾车、行走、进出楼宇、手持物品行进等，并全程被录制。场景中除常规动作外，还包含若干可疑行为（suspicious behaviors）。

异常说明：由于两次拍摄期间使用的设备种类繁多，部分相机出现故障，导致部分视频存在片段缺失。为保证各视频间的时间同步，我们在缺失时段添加了黑帧（black frames）。下文列出存在问题的视频及其缺失时长：
F1C3：缺失前66秒。
F1C5：缺失前2秒。
F1C8：缺失前3秒。
F1C13：缺失前10秒。
F1C15：缺失前1秒。
F1C19：缺失前1秒。
F2C1：视频存在加速问题且仅持续数秒，故未提供该视频。
F2C6：缺失时段为4:01至4:12以及4:25至4:28。
F2C16：缺失时段为5:15至5:26。

部分视频由移动设备录制，其像素分辨率（pixel resolution）低于1920×1080：
F1C3与F2C3：分辨率为1280×720。
F1C4与F2C4：分辨率为640×480。
F1C15与F2C15：分辨率为1280×720。
F1C20与F2C20：分辨率为1440×1080。

更多关于相机部署位置的详细信息可访问以下链接：http://ubee.enseeiht.fr/dokuwiki/doku.php?id=public:tocada

引用信息：T. Malon、G. Roman-Jimenez、P. Guyot、S. Chambon、V. Charvillat、A. Crouzil、A. Péninou、J. Pinquier、F. Sèdes 与 C. Sénac，《图卢兹校区监控数据集：场景、音轨、带重叠与非重叠视场的同步视频》，ACM 多媒体系统会议，2018年。

搜集汇总

背景与挑战

背景概述

Toulouse Campus Surveillance Dataset（ToCaDa）是一个用于监控研究的视频数据集，包含两个脚本化场景，每个场景提供25个时间同步的视频，涵盖重叠和不相交的视场。数据集提供了多种分辨率的视频文件以及音频和视觉注释，适用于取证和多源分析等应用。