TrackingLine

TrackingLine路径分割数据集共有三个子场景，分别来自第十二届“中国软件杯”大学生软件设计大赛智能四足机器狗电力巡检系统开发赛道(cnsoftbei)、第二十五届中国机器人及人工智能大赛机器人应用赛四足极速物流赛道(craic)和2023睿抗机器人开发者大赛物资运送赛道(raicom)。三个子场景的细节对比如下表。

|场景|训练集样本数量|验证集样本数量|合计数量|路径类型|路径颜色|背景颜色|
|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
|cnsoftbei|850|363|1213|窄，多曲线(大曲率)|黑色|白色，有不同颜色图案干扰|
|craic|500|214|714|宽、直线、直角弯|黄色|白色，无其他图案|
|raicom|1274|546|1820|宽，多曲线(小曲率)|黄色|白色，无其他图案|

我们使用宇树科技Go1四足机器人的下颚相机进行数据采集，标注后以7:3的比例将数据集划分为训练集和验证集，并以MS-COCO格式保存。三个子场景样例见下图。


The TrackingLine path segmentation dataset comprises three sub-scenarios sourced from three competition tracks: the Intelligent Quadruped Robot Dog Power Inspection System Development Track (cnsoftbei) of the 12th "China Software Cup" College Students Software Design Competition, the Quadruped High-speed Logistics Track of the Robot Application Competition (craic) at the 25th China Robot and Artificial Intelligence Competition, and the Material Transportation Track (raicom) of the 2023 RAICOM Robot Developer Competition.

A detailed comparative breakdown of the three sub-scenarios is provided in the table below:
| Scenario | Number of Training Samples | Number of Validation Samples | Total Samples | Path Type | Path Color | Background Color |
|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
| cnsoftbei | 850 | 363 | 1213 | Narrow, multi-curved (high curvature) | Black | White with interference from diverse colored patterns |
| craic | 500 | 214 | 714 | Wide, straight-line, right-angle turn | Yellow | White without additional patterns |
| raicom | 1274 | 546 | 1820 | Wide, multi-curved (low curvature) | Yellow | White without additional patterns |

Data was collected using the jaw-mounted camera of the Unitree Go1 quadruped robot. Following annotation, the dataset was split into training and validation subsets at a 7:3 ratio, and stored in MS-COCO format. Sample examples of the three sub-scenarios are illustrated in the following figure.