As a core sensor for autonomous driving and robotic perception, the accuracy of LiDAR data directly determines the system's ability to understand its environment. However, in practical applications, point cloud data rarely perfectly mirrors the physical world, with point cloud smearing being a common issue. Visually, this effect manifests as false point arrays extending from object edges or stretched shadows during high-speed motion. These artifacts not only distort geometric contours but may also mislead target recognition algorithms into misidentifying false smears as obstacles.

Causes of Energy Slicing in Geometric Optics

The fundamental physical cause of smearing in LiDAR stems from the fact that the laser beam is not an ideal, infinitely thin geometric line. After emission, the laser spot gradually diverges with increasing propagation distance due to the diffraction limit and collimation precision of the optical system.

This divergence, determined by the laser's wavelength and the waist diameter of the emitting optical system, forms an energy cluster with a specific area. When this energy cluster is projected onto the edges of two objects with depth differences during scanning, part of the spot falls on the nearer foreground target, while another portion may pass through the edge onto the farther background surface.

Image source: Internet

This phenomenon is known as the mixed-pixel effect at the sensor perception level. LiDAR's detection principle records the time difference between emitted pulses and received echoes. However, when the spot crosses an edge, the detector receives an energy superposition from two reflective surfaces within an extremely short time window.

If the distance between these two reflective surfaces is small, the echo signals experience severe overlap and distortion on the time axis. The sensor's internal timing circuit, whether using threshold detection or waveform sampling, faces a dilemma: should it record the foreground distance or the background distance?

When the system determines distance through weighted algorithms or centroid localization, the final ranging result neither belongs to the foreground nor the background but occupies a false position in between. In the generated point cloud, these false points resemble "tails" extending backward from object edges, with spatially continuous distributions—this is the geometric origin of smearing.

The laser beam's divergence angle is typically measured in milliradians. Even with extremely high collimation, the spot diameter may reach tens of centimeters beyond 100 meters. At this physical scale, the probability of the spot covering multiple depth surfaces increases significantly. This geometric ambiguity not only introduces distance measurement biases but also directly interferes with the accuracy of reflection intensity information.

Due to differing material reflectivities between foreground and background objects, minor variations in energy distribution ratios cause shifts in the received detection waveform's center—a phenomenon known as centroid drift. During scanning, as the spot moves completely from foreground to background, the measured depth values undergo a false transition from near to far, forming discrete, directional noise points in the point cloud. The severity of this optically induced smearing depends on multiple factors.

Additionally, the angular resolution of the optical system relative to spot size affects point cloud formation. If the radar's step angle exceeds the laser spot's coverage, the point cloud appears relatively sparse, but edge mixed-pixel points manifest as isolated noise.

In autonomous driving systems pursuing high-precision perception, extremely small step angles are often used to capture fine textures. This results in significant physical overlap between adjacent scan points. While this enhances perceived smoothness, it exponentially increases mixed-pixel occurrence. Without backend algorithms capable of identifying distance anomalies caused by energy slicing, a thin false surface forms at object boundaries, severely complicating environmental modeling.

Causes of Motion Displacement Along the Temporal Axis

Beyond geometric smearing from optical propagation, LiDAR's self-motion in dynamic scenes constitutes another major factor. This phenomenon, known as motion distortion in professional circles, arises from temporal asynchrony between point cloud data framing and sensor motion.

Typically, a complete LiDAR point cloud frame comprises tens of thousands of continuously emitted laser points accumulated over time. For mechanically scanning radars, a full 360-degree sweep takes approximately 100 milliseconds.

During this scan period, if the LiDAR-equipped vehicle or robot moves at high speed, turns sharply, or experiences severe bumps, the radar's absolute position and orientation continuously change while scanning each point. This means the coordinate system origins for the first and last points in the same frame have already undergone physical displacement.

If the processing system indiscriminately projects these points into a single static coordinate system, the point cloud exhibits obvious "deformation" or "stretching." This edge misalignment resembles photographic "camera shake," causing originally regular walls to appear tilted or cylindrical utility poles to transform into spiraling false trajectories. Such motion-induced false extensions constitute the second major physical source of point cloud smearing.

From a microscopic perspective, the laser pulse's round-trip flight time is extremely short (nanosecond scale), making individual point ranging results highly precise instantaneous values. Nevertheless, LiDAR does not capture global images instantaneously like cameras but operates through serial scanning. For rotating radars, the superposition of motion and scan vectors fundamentally alters local point cloud topology.

For example, when the radar moves forward, laser points scanning frontal objects appear closer than their actual positions due to the radar's approach, while points scanning rear objects appear farther. This systematic cumulative distance bias manifests as overall object contour drift or ghosting at the end of the scan cycle. For non-repetitive scanning solid-state LiDARs, though their scan trajectories are more complex than rotating types, motion distortion remains physically inevitable as long as frame acquisition requires time accumulation.

To address motion-induced smearing from temporal differences, high-frequency motion compensation mechanisms are essential. By fusing data from inertial measurement units (IMUs) and global positioning systems (GPS), the radar's subtle attitude changes during scanning can be monitored in real time.

Algorithms must assign precise timestamps to each laser point within a frame and calculate the sensor's pose matrix at that moment through interpolation. By reprojecting all points into a unified coordinate system at the frame's start time, deformation caused by self-motion can be effectively neutralized.

However, the effectiveness of this correction algorithm heavily depends on time synchronization precision between sensors. Microsecond-level synchronization errors or insufficient IMU sampling rates to cover high-frequency vibrations can still produce subtle, jagged fluctuations at point cloud edges. Though less spatially extensive than mixed-pixel effects, these fluctuations remain extremely difficult to eliminate when constructing high-precision maps, representing a form of dynamic smearing.

Electrical Bottlenecks at the Silicon Side

Shifting focus to LiDAR's internal electronic structure reveals that electrical signal processing directly and significantly modulates smearing phenomena. The LiDAR receiver is an ultra-sensitive system that must capture extremely weak photon signals reflected from hundreds of meters away.

To achieve long-range detection, designers typically assign extremely high gains to photodetectors (such as avalanche photodiodes, APDs) and front-end transimpedance amplifiers (TIAs). However, this high gain produces severe negative effects when facing nearby targets or high-reflectivity objects (e.g., traffic reflectors, vehicle mirrors): signal saturation.

When high-intensity reflected light momentarily strikes the detector, the resulting transient current may far exceed the circuit's linear dynamic range, causing the TIA to enter saturation lockup. Electrically, saturation severely clips and broadens the output voltage waveform. Originally nanosecond-wide narrow pulses experience significant elongation of their falling edges after passing through saturated circuits.

Since most ranging timing circuits capture signal arrival times via voltage thresholds, this pulse broadening directly causes timing delays. The timing circuit cannot distinguish whether this delay results from greater physical distance or circuit saturation. Consequently, dense false points often appear behind high-reflectivity objects in the point cloud, extending backward in depth—forming classic electrically induced smearing.

Furthermore, amplifier recovery from saturation to normal operation requires a physical process known as overload recovery time. During this "blind zone" before complete recovery, the circuit's baseline level drifts or becomes extremely sluggish in responding to subsequent signals.

If weak echoes from farther objects arrive during this period, they may be submerged in the lingering saturation artifacts or incorrectly discriminated. This electronic "hysteresis" effect not only produces axial smearing but may also cause complete loss of weak-signal targets near strong light objects.

To counteract these physical limitations, current high-performance radar chips employ multi-stage clamping techniques and adaptive gain control (AGC) circuits. Some TIA chips specifically designed for automotive environments can recover signals to normal levels within 10 nanoseconds after being impacted by 100-milliampere-level severe current surges, thereby minimizing measurement smearing caused by electrical saturation.

Beyond circuit saturation, the semiconductor properties of detectors themselves contribute to smearing. For Geiger-mode single-photon avalanche detectors (SPADs), once triggered into avalanche, internal charge clearance requires a certain "dead time."

In bright environments, random triggering by ambient photons intertwined with valid signal photons shifts the center of mass of timing statistical distributions. This manifests macroscopically as systematic distance measurement errors in the point cloud, appearing as blurred object surfaces or dispersed contour edges.

This hardware-level physical limitation-induced smearing follows strong regularities: stronger reflected energy correlates with longer smearing. Therefore, algorithms can compensate for such errors by analyzing each point's echo intensity. However, completely eliminating interference from silicon physical limits remains one of the most challenging areas in current LiDAR hardware design.

Environmental Media and Multipath Reflections

LiDAR's operating environment in the real world is far more complex than laboratory simulations, with environmental media properties constituting significant inducers of unstructured smearing. Under adverse weather conditions like rain, snow, or fog, the air becomes filled with microscopic water droplets, snow crystals, or particulates. When laser beams pass through these media, noticeable scattering occurs—primarily Mie scattering.

Part of the laser energy is reflected back to the receiver by these suspended particles. Although single-particle echoes are extremely weak, the cumulative energy from thousands of particles can trigger the sensor's detection threshold. These false points distributed between objects and the radar form diffuse, cloud-like smearing in space, significantly reducing the point cloud's signal-to-noise ratio.

More complex environmental interference arises from multipath reflections. When laser beams encounter highly reflective smooth surfaces like roadside mirrors, glass facades, or waterlogged road surfaces after rain, their propagation behavior shifts from diffuse to specular reflection.

When laser light strikes glass, part reflects while another part penetrates. If the reflected light subsequently hits another obstacle and returns to the radar, the system records the total "round-trip" flight time.

In the point cloud generation's projection logic, the system assumes straight-line propagation without knowing about the light's redirection. Consequently, the radar "paints" the actual obstacle behind the glass wall, creating mirror-image ghost points. When scan angles change slightly, this folded path continuously alters, producing a series of geometrically structured smearing points along the glass edge or behind it.

Among phenomena caused by high-reflectivity objects exists a special "halo" effect, somewhat analogous to lens flare in photography. When laser beams strike extremely reflective targets, they not only cause electrical broadening in the receiving channel but may also scatter excess energy internally within the optical system, interfering with adjacent detection pixels.

This lateral energy overflow significantly enlarges originally small objects in the point cloud, making them appear to "expand" in space. The combined action of such lateral and longitudinal displacements forms three-dimensional noise clusters in the point cloud that are difficult to remove through simple filtering.

Against environmentally induced smearing, multi-echo technology (Multi-echo) provides initial physical isolation. Echoes generated by laser beams encountering raindrops separate temporally from those generated by rear obstacles. Radars capable of simultaneously recording multiple echoes can use algorithmic logic to select the farthest one, thereby filtering out most rain and fog noise.

Furthermore, optical spatial filtering technologies are now being incorporated into advanced radar systems. These systems utilize minuscule apertures positioned on the receiving optical focal plane, enabling only collimated light beams at specific angles to pass through while effectively blocking the majority of off-axis stray light generated by atmospheric scattering. This process, known as "optical pre-filtering," not only enhances the signal-to-noise ratio but also eliminates false smearing points that arise from diffuse scattering at the source.

Concluding Remarks

The issue of LiDAR point cloud smearing is a multifaceted challenge that encompasses geometric optics, motion mechanics, semiconductor physics, and environmental science. A thorough understanding of the principles underlying mixed pixels, scan motion time differences, amplifier nonlinear responses, and multipath wave propagation mechanisms is crucial for effective preprocessing of high-quality point clouds and algorithm optimization. While current approaches, including high-frequency pose compensation, multi-echo logic, and deep learning filtering techniques, have substantially alleviated these problems, achieving even more precise removal of physical residues under extreme operating conditions is vital for advancing autonomous driving perception systems to a higher level.