Repository of camera trap data recorded during three pilot studies of the Amsterdamse Waterleidingduinen

Three camera trap data packages (https://camtrap-dp.tdwg.org/) of data collected as part of pilot studies carried out in the Amsterdamse Waterleidingduinen. The pilots were aimed at determining how different types of camera deployment (e.g. regular vs. wide lens, various heights, inside/outside exclosures) might influence species detections, and how to deploy autonomous wildlife monitoring networks. Two pilots were conducted in herbivore exclosures and mainly detected European rabbits (Oryctolagus cuniculus) and red fox (Vulpes vulpes). The third pilot was conducted outside exclosures, with the European fallow deer (Dama dama) being most prevalent. Across all three pilots, a total of 47,597 images were annotated using the Agouti platform. All annotations were verified and quality-checked by a human expert. A total of 2,779 observations of 20 different species (including humans) were observed using 11 wildlife cameras during 2021–2023. The raw image files (excluding humans), image metadata, deployment metadata and observations from each pilot are shared using the Camtrap DP open standard and the extended data publishing capabilities of GBIF to increase the findability, accessibility, interoperability, and reusability of this data. The data are freely available and can be used for developing artificial intelligence (AI) algorithms that automatically detect and identify species from wildlife camera images. The repository contains a data package in Camtrap DP format for each of the three pilots. Camtrap DP is an open standard for the exchange and archiving of camera trap data using a standardized data structure. Each data package consists of the following resources: · datapackage.json: Contains metadata about the data package and camera trap project from which the data originates. Describes taxonomic, temporal, and spatial extent. · deployments.csv: Table of individual camera trap deployments, detailing exact location and times active of each camera deployment. · media.csv: Table detailing every image in the data package. Lists the filenames and paths of images within the data package. · observations.csv: Table of observations of species (or lack thereof) derived from the images. · events.csv: Table linking observation events to media. · media folder: Folder containing a subfolder for each deployment, which contains the raw images from that deployment. Some additional notes, specific to these datasets: · The deployment table contains “deployment tags”, which specify extra information about the deployment, formatted as key:value pairs, separated by pipes (‘|’). Of particular interest for these datasets are the tags that state lens angle, specify habitat type and identify paired cameras (e.g. to assess differences in species detections between cameras with regular and wide lens, respectively). · In all three pilots, most observations are linked to sequences of images recorded within 120 seconds of each other. Hence, observations in these datasets are generally linked to an “event” (i.e. a sequence of images) rather than to an individual media file. We have added an events table to more easily link observation events and the media items that make up that event. This is an extension of the camera trap DP standard. · All annotations were verified and checked by a human expert, even in cases where an observation is listed as being made by an AI algorithm. · Whether or not an image is included in the data package is indicated by the ‘filePublic’ column in the media table. All raw images are included except for those where humans were detected. Images in which humans were detected have a ‘filePublic’ value of FALSE. Although the current location of these files within the Agouti platform (https://www.agouti.eu/) is recorded in the ‘filePath’ column, these files cannot be accessed. The ‘fileName’ of these filles is the original filename they possessed when uploaded to Agouti. · Where ‘filePublic’ is TRUE, the `filePath` given is relative to the root of the data package (e.g. ‘media/<deployment>’) and the `fileName` of the file is the current name of the file within the data package (‘<mediaID>.JPG’). More details about individual metadata fields in the Camtrap DP format can be found on https://camtrap-dp.tdwg.org/.本数据集包含3套相机陷阱（camera trap）数据集包（https://camtrap-dp.tdwg.org/），数据采集自阿姆斯特丹水防护林区（Amsterdamse Waterleidingduinen）开展的系列试点研究。本次试点的核心目标为探究不同相机布设方案（如常规镜头与广角镜头、不同架设高度、围栏内外布设等）对物种检出效果的影响，以及如何搭建自主野生动物监测网络。 其中2项试点在草食动物围栏内开展，主要检出欧洲兔（*Oryctolagus cuniculus*）与赤狐（*Vulpes vulpes*）；第3项试点在围栏外开展，优势物种为欧洲黇鹿（*Dama dama*）。 三项试点总计使用Agouti平台（Agouti）标注了47597张图像，所有标注均经人工专家审核并完成质量校验。2021—2023年间，依托11台野生动物相机共记录到包含人类在内的20个物种的2779条观测记录。 本数据集的原始图像文件（不含含人类的图像）、图像元数据、布设元数据以及各试点的观测数据，均采用Camtrap DP（Camtrap DP）开放标准，并借助全球生物多样性信息机构（GBIF）的扩展数据发布能力进行共享，以提升该数据集的可发现性、可访问性、互操作性与复用性。本数据免费开放，可用于开发从野生动物相机图像中自动检测和识别物种的人工智能（AI）算法。 本数据仓库包含3个试点各自对应的Camtrap DP格式数据集包。Camtrap DP是一套采用标准化数据结构的相机陷阱数据交换与归档开放标准。每个数据集包包含以下资源： 1. datapackage.json：包含数据集包及所属相机陷阱项目的元数据，描述了分类学范围、时间范围与空间范围。 2. deployments.csv：单台相机陷阱布设信息表，详细记录了每台相机的精确布设位置与活跃时段。 3. media.csv：数据集内所有图像的详细信息表，列出了数据集中图像的文件名与存储路径。 4. observations.csv：由图像衍生的物种（或无物种）观测记录表。 5. events.csv：关联观测事件与对应图像的表格。 6. media文件夹：包含每个布设点位对应的子文件夹，存储该布设点位的原始图像。 针对本数据集的额外说明如下： 1. 布设信息表包含‘布设标签’，以键值对格式、通过竖线（|）分隔，用于标注布设的额外信息。本数据集的核心关注标签包括：镜头角度、生境类型标识以及配对相机标识（例如用于对比常规镜头与广角镜头相机的物种检出差异）。 2. 三项试点的绝大多数观测均关联到120秒内录制的连续图像序列，因此本数据集的观测通常关联至‘事件’（即图像序列）而非单张媒体文件。我们新增了events表以简化观测事件与组成该事件的媒体文件之间的关联操作，这属于对相机陷阱DP标准的扩展。 3. 即便观测记录标注为通过AI算法生成，所有标注均经人工专家审核校验。 4. 媒体表中的filePublic字段用于标识图像是否包含在数据集中：除检测到人类的图像外，所有原始图像均被纳入。含人类的图像filePublic字段值为FALSE。尽管Agouti平台（https://www.agouti.eu/）中这些文件的当前存储路径记录于filePath字段，但无法访问这些文件。这些文件的fileName为上传至Agouti时的原始文件名。 5. 当filePublic为TRUE时，给出的filePath为相对于数据集包根目录的相对路径（例如media/<deployment>），文件的fileName为数据集中当前使用的文件名（<mediaID>.JPG）。 6. 关于Camtrap DP格式中各元数据字段的更多详细信息，可访问https://camtrap-dp.tdwg.org/查阅。