"Detailed Explanation of LiDAR Algorithm for Intelligent Driving" 1. What is LiDAR

戈壁滩上的辉煌

"Detailed Explanation of LiDAR Algorithm for Intelligent Driving" 1. What is LiDAR [Copy link]

Basic principle:

Laser radar, or Light Detection and Ranging (LiDAR), is based on the core concept of light detection and ranging. Taking the on-board mechanical laser radar as an example, its operating mechanism is simple and efficient: the laser emits pulsed lasers, which accurately touch the target surfaces such as the ground, vegetation, vehicles, road facilities and buildings, and then scatter, and some light waves are reflected and return to the laser radar receiver. Using the principle of laser ranging, the precise distance between the laser radar and the target point is calculated instantly. As the laser rotates horizontally, this measurement process continues, fully capturing the laser point data of all targets around the vehicle. Through advanced data imaging processing technology, these laser point data are converted into a three-dimensional point cloud map of the surrounding environment, realizing accurate depiction and perception of the surrounding environment.

Since the birth of the first laser in 1960, LiDAR has been widely used in the military field, and then entered the scientific research mapping, weather forecasting, industrial measurement and robotics industries. With the development of unmanned driving technology, LiDAR has been promoted in the perception scheme of unmanned vehicles. China, the United States and Europe have promoted intelligent driving technology, and the LiDAR industry has entered a period of rapid development and has been applied in many commercial fields.

The main categories are as follows:

Mechanical scanning LiDAR technology is mature, and there are a variety of line count products to choose from, but the high cost and large size make it difficult to meet the needs of mass-produced vehicles. MEMS galvanometer, transmission prism and rotating mirror LiDARs are gradually being adopted by automobile manufacturers. Flash and OPA LiDARs are considered to be the technical trends of future automotive LiDARs, but they are currently less used.

Characteristics of different perception sensors:

Core applications in intelligent driving systems

Based on the diversified functions of the internal algorithms of the intelligent driving system, the system can be refined into key subsystems such as perception, simultaneous localization and mapping (SLAM), prediction, decision-making, planning and control. LiDAR, with its excellent three-dimensional environment detection capabilities and highly sensitive reflection intensity recognition of target material differences, has become an indispensable component of the perception and positioning system of intelligent vehicles. In the field of perception, LiDAR has demonstrated its unparalleled strength. It is seamlessly integrated into the various functional modules shown in Figure 1-7, accurately identifying road layout, obstacle locations, dynamic targets and safe driving areas, providing intelligent vehicles with comprehensive and detailed surrounding environment perception capabilities.

在工程实践中，3D目标检测模块专注于识别车辆、行人、骑行者等核心目标，而小障碍物则依赖聚类功能有效识别，此举旨在缩减标注成本并简化深度神经网络的检测复杂度。融合聚类与3D检测功能，显著增强了感知算法的稳健性，有效应对奇异物体引发的漏检问题，确保智能车安全行驶。鉴于驾驶环境的复杂性，如目标遮挡与点云噪声，单帧检测常难以精准捕捉目标信息。因此，我们采用多帧融合策略，优化目标信息的估计精度。同时，为实现高效的目标管理，系统为检测到的行人、车辆等分配唯一ID，并估算其速度，为后续轨迹预测奠定基础。这一过程由先进的多目标跟踪模块精准执行，如图1-10所示，在高速场景下，通过连续三帧点云实现了目标跟踪与ID精准分配。驾驶过程中的道路感知同样重要，包括道路边界与车道线信息的准确获取，这依赖于路沿检测与车道线检测模块的高效协作。值得注意的是，早期由于激光雷达线数限制及数据稀疏性，相关公开研究较为有限。但随着技术进步，这一领域正逐步拓展，为智能驾驶提供更为全面、精准的感知支持。

The application functions of vehicle-mounted LiDAR include laser point cloud, clustering algorithm, instance/panoramic segmentation model and L-Shape Fitting algorithm for the target, which can realize target size, orientation, tracking, target management and speed estimation. At the same time, the lane line, curb and drivable area of the background point cloud can be segmented and calculated. In the field of intelligent driving, the application of LiDAR in SLAM function includes intelligent driving functions such as memory parking and automatic assisted navigation driving, which can realize all-weather, high-precision and real-time positioning of the vehicle. In order to make up for the shortcomings of GPS, many R&D teams use a combination of high-precision maps and high-precision positioning. By matching LiDAR point clouds and images with high-precision maps, combined with GPS signals, the absolute position estimation of the vehicle can be realized.

On-board LiDAR is widely used in Robotaxi and driverless trucks. Robotaxi is usually equipped with LiDAR, as is TuSimple's driverless truck. The perception technology routes of vehicle manufacturers are mainly divided into two directions: vision-led and LiDAR-led. Tesla uses a vision-led perception solution, using a combination of "camera + millimeter-wave radar". Tesla's AutoPilot HW2.0 system is equipped with 8 cameras, 1 millimeter-wave radar, and 12 ultrasonic radars. LiDAR is being used more and more widely in intelligent driving, and many car companies have begun to use LiDAR as the main sensor. The price of LiDAR is gradually decreasing, and it may be cheaper in the future. Some models such as ET7, M7, Xiaopeng P5, Ideal L9, etc. are equipped with LiDAR. With the increasing application of LiDAR in the automotive industry, the demand for LiDAR algorithm engineers is also increasing.

吾妻思萌

I have been taught, and I am learning.