Feb 6
Data Annotation for High-Precision Lane Line Recognition in Autonomous Vehicles
With the rise of autonomous vehicles, the demand for accurate lane line recognition has increased significantly. Lane line recognition is a crucial component of autonomous driving, as it helps the car accurately identify the road and make informed decisions for safe navigation.
Camera technology plays a vital role in the perception aspect of autonomous driving. It serves as the vehicle's " eyes, " collecting information on surrounding objects such as vehicles, pedestrians, and road signs.
To ensure the safety of autonomous vehicles, it’s essential to have high-precision lane line recognition. Currently, the quality of AI data used for autonomous driving solutions varies greatly, and high-quality data can significantly improve the efficiency of artificial intelligence. In contrast, low-quality data can hinder its evolution.
This case study highlights a data annotation process for lane line recognition. The labeling types include:
- Segmentation annotation: using polygon shapes to label road signs such as speed bumps and inverted triangle signs.
- Polyline annotation: labeling yellow and white lane lines on the road.
- General requirements for the annotation process include labeling all relevant objects in the visual field, using lines and polygon shapes, and ensuring accuracy with no missing or incorrect labeling.
Labeling tasks and labeling categories
Labeling requirements
1 Segmentation
- The edges of the polygon should match the outfit of the subject.
2 Lane lines
- Draw a line along the centerline of the lane.
- If a lane line has both a dashed line and a solid line, select the one with a longer length to label the category.
- The labeled line should be a smooth curve.
- If the opposite lane is not isolated by fences, cement piers, etc., the lane line should be labeled; Lanes on roads separated by fences and concrete walls don’t need to be labeled.
- There shall be no missing or wrong labeling. Lines that are less than 40 pixels in length in the image don’t need to be labeled.
- If a line has both visible and occluded parts, the occluded parts should be labeled.
Specification
1 Segmentation
- White and yellow solid lines: You need to draw the whole line with a polygon.
- White and yellow dashed lines: Each section is marked with a polygon.
- Stop line: The whole line needs to be labeled with a polygon.
- Zebra crossing: Draw each one with a polygon. If you can’t distinguish each one clearly, draw multiple lines with one polygon.
- Road arrow: need to draw a road arrow with a polygon.
- Horizontal deceleration marking: draw each one with a polygon. If you can’t distinguish each one clearly, draw multiple lines with one polygon.
- Inverted triangle sign: Each triangle should be labeled with a polygon.
- Road text: each character should be marked in a polygon. The road text should not be marked too closely and should be drawn in a quadrilateral.
- The numbers should be marked in one polygon if they are as a whole. For example, 35 should be in one polygon, instead of using two polygons to separate 3 and 5.
- Diamond signs: Each diamond is labeled with a polygon.
- No stopping zone marking: draw with multiple polygons.
- Guidelines: draw each one with a polygon. If you can’t distinguish one by one, draw multiple lines with one polygon.
- Parking line: draw with multiple polygons.
- Parking space text: Each text character is marked with a polygon.
2 Label lane lines
Lane line annotation is mainly to label the common yellow-white solid and dashed lines and stop lines on the road, to ensure that they are centered. The tag and the line lengths are correct.
Output
End
Outsource your data labeling tasks to ByteBridge, you can get high-quality ML training datasets cheaper and faster!
- Free Trial Without Credit Card: you can get your sample result in a fast turnaround, check the output, and give feedback directly to our project manager.
- 100% Human Validated
- Transparent & Standard Pricing: clear pricing is available(labor cost included)
