航空智能监测基准红外数据集(BIRDSAI)

监测保护区以遏制偷猎等非法活动是一项艰巨的任务。随着无人机(UAV)和TIR摄像机等传感器的进步，实时数据采集变得更加容易，因为这些传感器允许在盗猎通常发生的夜间进行监视。然而，准确和快速处理大量产生的TIR数据仍然是一个挑战。航空智能监测基准红外数据集(BIRDSAI，发音为“鸟眼)是一个包含南部非洲动物和人类夜间图像的长波热红外(TIR)数据集。该数据集允许测试用真实和合成的视频自动检测和跟踪人类和动物，以保护现实世界中的动物。
有48个真实的空中国际公路货运视频和124个合成的空中国际铁路货运视频(使用AirSim生成)，总共分别为62k张和100k张图像。这些视频中为每种动物和人类提供了跟踪信息。我们会将这些信息分解为动物或人类的标，并在可能的情况下提供物种信息，包括大象、狮子和长颈鹿的信息。我们还为每个限界框提供了噪音和遮挡的信息。
在提供的训练集中，有两个文件夹，一个用于模拟数据(TrainSimulation)，一个用于真实数据(TrainReal)。每个文件夹都包含用于每个视频的注释.csv文件(annotations)和每个视频中的单个.jpg帧(images)的文件夹。
在“TrainSimulation”中的“images”文件夹中，每个视频都有一个文件夹;除了.pg红外图像外，这些zip文件还包含Airsim提供的红外.png、RGB和分割图。

Monitoring protected areas to curb illegal activities such as poaching is a daunting task. Advancements in sensors including unmanned aerial vehicles (UAVs) and thermal infrared (TIR) cameras have facilitated real-time data collection, as these sensors enable surveillance during nighttime, when poaching incidents typically occur. However, accurately and rapidly processing the large volumes of generated TIR data remains a substantial challenge.

The Benchmark Infrared Dataset for Aerial Intelligence Surveillance (BIRDSAI, pronounced "bird-eye") is a long-wave thermal infrared (TIR) dataset containing nighttime images of animals and humans in Southern Africa. This dataset enables testing of automated detection and tracking of humans and animals using both real and synthetic videos for real-world wildlife conservation applications.

The dataset includes 48 real aerial international highway cargo videos and 124 synthetic aerial international railway cargo videos (generated using AirSim), with a total of 62,000 and 100,000 images respectively. Tracking annotations are provided for each animal and human across all videos. We categorize these annotations into animal or human classes, and provide species information where feasible, including for elephants, lions, and giraffes. We also supply noise and occlusion information for each bounding box.

The provided training dataset contains two folders: one for simulated data (TrainSimulation) and one for real data (TrainReal). Each folder includes a .csv annotation file for each video, alongside an `images` folder that stores individual .jpg frames extracted from each video.

In the `images` folder within `TrainSimulation`, each video has its own subfolder. In addition to the .jpg infrared images, these zip files also contain infrared .png, RGB, and segmentation maps provided by AirSim.

搜集汇总

背景与挑战

背景概述

BIRDSAI是一个用于航空智能监测的热红外数据集，专注于夜间动物和人类图像的自动检测与追踪，以支持反偷猎保护。数据集包含48个真实视频和124个合成视频，分别提供62,000张和100,000张图像，结合真实与合成数据用于AI模型训练和测试。