How Adverse Weather Conditions impact on LiDAR Detection


Table of Contents

LiDAR point cloud data on rain and fog

For a long time, weather challenges have been the bottleneck for ADS, and many efforts have been made to address it. In fact, meteorology has long been studying the relationship between severe weather and road safety. Perry et al. reviewed the hazards of wet slippery roads, reduced road visibility, impacts on driver decision-making processes, accidents caused by weather, and international efforts to promote road safety long before autonomous driving cars attracted much attention in the market.

Self-Driving Vehicle Architecture: Adverse Weather Agnostic
Explore an architecture designed for self-driving vehicles, agnostic to adverse weather conditions. Red blocks represent weather-related modules, while blue arrows denote relationships among weather, perception, and sensing modules. Gray arrows illustrate connections among autonomous driving system (ADS) modules, including external weather factors like wind and wet road surfaces. This survey focuses mainly on the area enclosed in the dashed rectangle.


Over a decade ago, R.H. Rasshofer et al. attempted to analyze the impact of weather on automotive LiDAR systems. They proposed a method that simulated the light return signal measured by reference LiDAR under real weather conditions, which was developed prior to actual testing and artificial environment synthesis target simulation. Signal parameters such as pulse shape, wavelength, and power level were replicated, and the impact of weather was presented through analysis. However, considering that the real world is not static and synthetic targets cannot be exhaustive, this method is no longer reliable enough for ADS. Nevertheless, they were one of the pioneers in dealing with adverse weather.

Each phenomenon’s impact on sensors:

0 – Negligible: Almost negligible impact;

1 – Slight: Impact that hardly causes detection errors;

2 – Minor: Impact that causes minor errors in special situations;

3 – Moderate: Impact that causes perception errors up to 30% of the time;

4 – Severe: Impact that causes perception errors over 30% but less than 50% of the time;

5 – Very severe: Noise or blockage causing false detection or detection failure;

Some key factors, such as measurement range, measurement accuracy, and point density, may be affected by weather conditions, affecting the normal operation of autonomous driving vehicles. Since the concept emerged, people have tested and verified LiDAR or the entire AV mode under adverse weather conditions, whether in artificial environments such as fog chambers or in real-world scenarios such as Scandinavian snowfields, or even in simulated environments.

Multiple 3D LiDAR Brands: Velodyne, Hesai, RoboSense, Ouster, and More
Explore their performance under adverse weather, including Velodyne, Hesai, RoboSense, Ouster, and others.

1. Rainy and Foggy Weather Impacts on LiDAR Detection

Fersch et al.’s research shows that rain typically doesn’t affect LiDAR and AV systems with small-aperture receiver LiDAR sensors, except in heavy downpours. Raindrops interacting with the laser beam cause minimal power loss, usually below 10%. Wetting the transmitter window can reduce signal power, with smaller droplets having a greater impact. However, condensation near the dew point can lead to significant power loss. Rainfall can also affect the accuracy and integrity of point cloud data, beyond just signal power levels.

LiDAR point cloud data on rain and fog
LiDAR point cloud data on rain and fog

We can see from the LIBRE dataset conducted by Carballo et al. . The point cloud of LiDAR displayed disappointing results due to fog, rain, and humid conditions in Figures 3(b) and 6(c). The artificial rain generated in the fog chamber, the Japan Automobile Research Institute (JARI) weather experimental equipment, produced a new problem in this case, namely, most LiDAR detected water in the sprinkler as shown in Figure 3(d), and the vertical cylinder made the point cloud more chaotic. More than a decade ago, when researchers were still struggling to stabilize visibility control to provide better test environments, fog chambers had come a long way. However, before better replication systems are available, real weather testing may not be completely replaced by fog chambers, otherwise, the efficiency of such research may be affected.

2 Snow Impacts on LiDAR Detection

Unlike rain, snow consists of solid objects, snowflakes, which easily freeze together to form larger solids, causing false detections or obstructing the line of sight of LiDAR. Since snowfield test sites (such as fog chambers) are not easy to enter and there are obvious dangers in driving on snow, tests on snow effects are rare.

"LiDAR Point Cloud Data: Snow Weather Analysis
Explore the analysis of LiDAR point cloud data under snow weather conditions.

Jokela et al. tested the performance of AVs LiDAR detection under snow conditions in Finland and Sweden, mainly focusing on snow vortices caused by front cars. Figure 4(a) shows that for typical VLP-16 LiDAR (Velodyne’s Puck LiDAR), except for some noise points near the sensor, almost no snow is detected, but gaps are generated in the LiDAR point cloud due to the front and rear views of the car itself. LiDAR from different manufacturers may have different performances on the same matter. Figure 4(b) shows that RS-LiDAR-32 has high-density clouds caused by snow around the sensor, similar to snow vortices. Therefore, it can be said that snow or rotating snowflakes in the atmosphere may cause abnormal point clouds of LiDAR and shorten the detection distance.

More importantly, LiDAR such as Velodyne VLP-16 used in this work has a minimum operating temperature of -10°C. In cold environments that are not uncommon in the northern hemisphere, the delay measured by LiDAR increases by about 6.8 ns when the temperature changes greatly, such as from extremely cold (-20°C) to extremely hot (+60°C) environments, which widens the LiDAR ranging beyond 1 meter and reduces near-field accuracy, not to mention the sensitivity and distance measurement of photodetectors.

3 Other Impacts

Currently, most LiDAR manufacturers use lasers with a wavelength of 905 nm. However, LiDAR upgrades have always been a focus of attention in the research community. Kuttila et al. [76] proposed the use of 1550 nm LiDAR to overcome foggy conditions because it allows higher laser power to be emitted at this wavelength while ensuring eye safety. Before we determine its feasibility, it is necessary to raise two key design considerations in LiDAR selection: eye safety and environmental interference. Most civil or commercial LiDARs are used in environments where human eyes are exposed, so according to the International Laser Product Safety Standard (IEC 60825), the infrared laser of LiDAR must not exceed the maximum permissible exposure (MPE) or cause any damage to the retina-Class 1.

Therefore, the choice of laser wavelength is almost narrowed down to two options: 800 nm – 1000 nm and 1300 nm – 1600 nm. That is why there are currently choices of 850 nm 3, 905 nm 4, and 1550 nm 5 wavelengths for LiDAR manufactured for autonomous driving cars, which also belong to the low solar irradiance wavelength range, which helps suppress the low SNR of the signal receiver due to low environmental light. The extinction coefficient of 1550 nm wavelength is relatively large (Figure 8), which can absorb a large amount of light in the crystalline lens or vitreous body of the eye, allowing a higher power to be emitted than 905 nm under the condition of ensuring eye



Perception and Sensing for Autonomous Vehicles Under Adverse Weather Conditions: A Survey


Request User Manual

Request Datasheet