1. Definition:
A point cloud is a collection of points acquired in 3D space and output by LiDAR. By recording the reflection of laser pulses off object surfaces, it translates the geometric shapes of the physical world into a massive set of discrete spatial coordinate points. It serves as the core foundational data for environment perception, localization, and high-definition map construction in autonomous driving.

Basic Components:
Each individual “point” carries multi-dimensional physical information, typically including the following core attributes:
(1) Spatial Coordinates (x, y, z): Establish the 3D geometric profile of the target, forming the basis for path planning and obstacle avoidance.
This figure illustrates how the raw spherical coordinates of LiDAR (range r, elevation angle θ, azimuth angle φ) are converted into Cartesian coordinates (x, y, z) of the point cloud, thereby defining the precise position of each point in 3D space.

(2) Reflection Intensity: Reflects material reflectivity. It is used to identify traffic markings and road signs, and to distinguish different road surface materials.
Through the stark contrast between green high-reflection areas (such as crosswalks and lane markings) and blue low-reflection areas (such as asphalt pavement), this figure intuitively demonstrates how the reflection intensity of a LiDAR point cloud reflects material differences on target surfaces.

(3) Timestamp: Records the precise acquisition time of each point. It is used for multi-sensor fusion and motion compensation to correct point cloud distortion during high-speed travel.
(Image source: https://inertiallabs.com/wp-content/uploads/2024/09/Screenshot-2024-09-10-162634.jpg)
The figure above shows a LiDAR point cloud colored by GPS Time, clearly displaying the time series of point-by-point scanning and intuitively demonstrating the core role of timestamps in motion compensation and multi-sensor fusion.

(4) Echo Attributes: Records the multiple reflection characteristics of pulses. It is used to achieve “see-through” perception in vegetation, rain, fog, and glass scenarios, filtering environmental noise and enhancing safety in extreme weather.
(Image source: https://inertiallabs.com/wp-content/uploads/2024/09/Screenshot-2024-09-10-162620.jpg) This figure illustrates the process of multiple echoes generated successively at the canopy, branches, and ground when a laser pulse penetrates vegetation, clearly demonstrating the physical mechanism of multi-echo recording.

3. Key Dimensions for Evaluating Point Cloud Quality
(1) Probability of Detection (PoD):
Refers to the probability that a LiDAR measurement forms a valid point, which can be calculated as the ratio of the number of valid points in the point cloud to the theoretical number of detected points.

“Valid points” refer to points in the LiDAR point cloud that correspond to real objects at actual spatial locations.
“Theoretical number of detected points” refers to the total number of points that should theoretically cover the target object, based on the angular resolution and scanning range of the LiDAR.

(2) False Point Rate
Refers to the ratio of the number of false points to the theoretical number of detected points.

“False points” refer to points in the LiDAR point cloud that correspond to locations in real space where no actual object exists.