10分钟猪跟踪视频数据集
收藏资源简介:
该数据集由拉瓦尔大学动物科学系创建,包含10分钟内猪的跟踪视频,共21个不确定的实体识别标识。数据集旨在用于长期多目标跟踪,特别是牲畜行业的应用。该数据集的创建过程涉及使用喂食站获取牲畜的实体识别标识,并将其与跟踪算法相结合。数据集的应用领域包括动物行为分析、牲畜管理和视频监控等,旨在解决长期多目标跟踪中目标身份切换和丢失的问题。
This dataset was developed by the Department of Animal Science at Laval University. It contains pig tracking videos spanning a 10-minute period, with a total of 21 uncertain entity recognition identifiers. The dataset is designed for long-term multi-object tracking applications, particularly for use in the livestock industry. The creation process of the dataset involved utilizing feeding stations to acquire entity recognition identifiers for livestock, and integrating these identifiers with tracking algorithms. The application areas of this dataset include animal behavior analysis, livestock management, video surveillance and other related fields, with the goal of resolving the problems of target identity switching and loss encountered in long-term multi-object tracking.
数据集概述
数据集名称
Uncertain Identity Aware Tracking
来源会议
CVPR2024 CV4animals workshop
数据集内容
- 数据集位于dataset文件夹内
- 包含猪只追踪相关数据
主要文件
- Bytetrack:包含获取追踪器转移矩阵、添加观测值和执行HMM追踪的代码
- forwardBackward.py:包含执行HMM追踪的代码
- ATQ.py:包含添加标识符信息的代码,可使用喂食器数据或人工生成数据
实验复现
生成访问测试
bash performance_test.py artificial_visits
真实喂食器数据测试
bash performance_test.py feeder
模型基准测试
bash performance_test.py tracker_test
评估指标
- 可通过Bytetrack/MOT_metric_evaluation/tracking_evaluation.ipynb笔记本可视化MOTA、IDF1、切换次数等指标
示例展示
HMM基于追踪框架修复追踪器身份切换的示例(猪只RWID 4818)




