Three methods and four applications of 3D point cloud annotation

What is a 3D point cloud?

3D point cloud is a method of representing the three-dimensional world with point cloud. It can be imagined as materializing three-dimensional objects and using multiple data points to represent an object.

Point cloud data is generally obtained by 3D scanning equipment such as lidar to obtain information of several points in space, including XYZ position information, RGB color information and intensity information, etc., which is a multi-dimensional complex data collection.

3D point cloud annotation method

3D point cloud labeling refers to the use of data collected by lidar for frame selection and labeling, which is used for training artificial intelligence models such as computer vision and driverless driving. There are currently three mainstream 3D point cloud labeling methods.

1. 3D point cloud continuous frame labeling: 3D point cloud continuous frame labeling is a widely used data processing type in autonomous driving scenarios, which requires high three-dimensional space perception capabilities and multi-frame collaborative processing capabilities.

2. 3D point cloud semantic segmentation annotation: point cloud semantic segmentation can select the marked object with a cube frame, and dye the point cloud into the same color, accurately identify pedestrians, cars and other objects, and help vehicles understand the road environment. Semantic Segmentation Annotation Prices and Tools (Graph Table)

Jinglianwen Data: “32,000 Portrait Semantic Segmentation Dataset” helps autonomous driving technology

3. 3D point cloud image annotation: 3D point cloud image annotation data is the basic training data of unmanned driving technology. 3D point cloud image annotation is to mark out the target object through the 3D frame in the 3D image collected by the lidar. Including vehicles, pedestrians, advertising signs and trees etc. What is Image Annotation? (easy to understand guide)


3D point cloud annotation application scenarios

3D point cloud can keep the original geometric information in 3D space without any discretization, which also makes its social application gradually deepen.

1. Self-driving cars:

Through the 3D point cloud semantic segmentation technology, the road environment point cloud data can be segmented and simultaneously positioned and mapped, and objects such as pedestrians and cars can be identified, which is most useful for self-driving cars.

2. Railway scene detection

3D point cloud technology can accurately identify foreign objects on the railway, and is not easily affected by weather and the environment; foreign object intrusion detection in important areas such as station platforms and tunnel entrances can effectively ensure the safety of high-speed railway operation, and lidar is not easy Affected by weather and environment, point cloud technology is widely used.


3. AR virtual reality

The use of 3D point cloud semantic segmentation technology allows people to experience virtual 3D scenes through AR glasses. It can effectively reflect the content of the real world, and can also promote the display of virtual information content.

4. AI artificial intelligence

In the artificial intelligence (AI) system, computing power, algorithms, and data are the three major elements of artificial intelligence evolution, which respectively undertake the role of artificial intelligence in infrastructure capabilities, work guidance methods, and algorithm (evolution) basis. Computing power is the ability of technical facilities, algorithm is the working method, and data is the basis for optimizing the algorithm. In other words, the first two are equipment and capabilities, and data is the knowledge material that can be learned by artificial intelligence

The 3D point cloud images generated by lidar can be used for automatic driving system training after annotation. With the improvement of performance, the amount of training data required by automatic driving technology is almost exponentially increasing. Common annotation methods include 3D rectangular frame annotation, 3D Point cloud semantic segmentation, point cloud continuous frame labeling, 2D3D fusion labeling, etc. JLW Technology’s self-built labeling platform can provide comprehensive tool support for point cloud data labeling.


Relying on a large amount of labeling experience accumulated in the field of intelligent driving, recently, Jinglianwen Technology has won the 3D laser point cloud labeling project for autonomous driving from a well-known manufacturer through trial labeling with extremely high labeling efficiency and accuracy. It reflects JLW’s strong technical support ability and project management ability.