A continuous video frames dataset of goose for computer vision

As research on geese is currently in a relatively new field, there is a lack of suitable publicly available datasets. The data were obtained from a free-range goose farm in Lubei Village, Xiaoyi City, Shanxi Province. In order to obtain continuous image data of geese in their natural environment, video data were collected from the breeding environment of free-ranging geese. The videos were captured in a top-down perspective at a height of 1.8 metres above the ground. The main data acquisition device was a Huawei HONOR 60, acquired on 10 May 2023, and the videos obtained included at least two types of disturbances: weed-filled environments and shadows from trees. Introducing these complexities ensures a realistic simulation of the video capture scene in a natural environment. Once the video data were acquired, the captured video was edited into images every 2 seconds. Images that are morphologically difficult to recognise, have unclear features or are too blurred are discarded, so that the majority of the images obtained are clear. Finally, a continuous video frame dataset is constructed.

当前鹅类研究尚处于较新的领域，适配的公开可用数据集较为匮乏。本数据集的数据采集自山西省孝义市鲁北村的一家散养鹅场。为获取自然环境下鹅群的连续图像数据，研究人员从散养鹅的养殖环境中采集了视频数据。视频采用俯视视角采集，拍摄高度距地面1.8米。本次数据采集的主要设备为华为荣耀60（Huawei HONOR 60），采集时间为2023年5月10日；所获取的视频包含至少两类干扰场景：杂草丛生的环境以及树木投射的阴影。引入这类复杂干扰因素，可实现对自然环境中视频采集场景的真实模拟。完成视频数据采集后，研究人员以每2秒一帧的频率将采集到的视频剪辑为图像。将形态难以识别、特征模糊或画质过度模糊的图像剔除，最终得到的图像大多清晰可用。最终构建得到连续视频帧数据集。