Object Recognition-Traffic Signs Annotation Case Study in Automatic Driving
The automatic recognition technology of road traffic signs is very important in autonomous driving. In realistic scenarios, interference factors such as light intensity, obstruction, and shooting angle often bring challenges to signage recognition.
From the perspective of the research direction of AI technology, supervised learning based on training samples with clear labels or results is still a major model training method, whether in traditional machine learning or deep learning.
To make autonomous vehicles more “intelligent”, autonomous driving applications need to form a commercial closed loop in different vertical landing scenarios, which needs the support of massive and high-quality real road data.
Let’s take a look at a traffic sign labeling case.
Specification of Labeling Categories
1.1 Annotation of traffic signs: Traffic signs can be divided into speed limit signs and non-speed limit signs.
Non-speed limit signs include the common weight limit signs, high limit signs, width limit signs, axle load limit signs, warning signs, prohibition signs, gantries, big square signs, information signs, separators, and other types, the classification is shown in the following table.
1.2 Maximum speed limit signs
The maximum speed limit signs include two types with big differences in characteristics. One is the common unchangeable signs, which are characterized by a round board, with a red circle, white background, and black words. The other type is changeable signs, which are characterized by electronic signs and variable values. But the core of the maximum speed limit signage is the value. For example, the value of 50 is always with the maximum speed limit of 50km/h.
1.3 Minimum speed limit signage
This type of sign has apparent features, such as a round board, blue bottom, white words, and flat bottom.
1.4 Remove the maximum speed limit signage
This type of sign has apparent features, such as a round board, black circle, black words, and four or five parallel black lines.
1.5 Non-speed limit signage
Weight limit, height limit, width limit, yielding, no-long-time parking, no vehicle parking, no driving, yielding for parking, and other prohibition categories are non-speed limit signs, as follows:
- Information Signs
- Prohibition Signs
- Axle Load
- Prohibition Signs
Scenario categories are annotated as follows:
2.1 Time of Scenarios
2.2 Areas of Scenarios
2.3 Weather of Scenarios
3.1 Annotated objects
All image traffic signs in the listed scenario should be labeled.
3.2 The borders of the box have to closely match the outline of the target
All types of signs shall be labeled in a clockwise direction. With the upper left corner as the starting point and serial number as 0, label the coordinate of signs in the whole image.
1） No annotation is required when the minimum border length is less than 10 pixels.
2） When the length and width are less than 15 pixels and the sign cannot be identified, no annotation is required.
3） If the signs are partially occluded, the invisible part should be labeled.
- The borders of the box should be close to the 4 edges of the target (within 3 pixels).
- There is no need to label an image that cannot be identified by the naked eye.
- The target of traffic signs is small, and it is likely to be unnoticed. Attention should be paid to the roadside and hidden corners of intersections.
- If the side angle of the traffic sign or the turning angle is too large, do not label it.
- If the length-breadth ratio is greater than 1.6, do not label.
- No need to label the traffic signs that are faded or seriously deformed.
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)