LiDAR Signal Enhancement for Accurate Detection

Laser radar point cloud denoising method using genetic algorithms to automatically determine optimal filter parameters for improved noise removal while preserving original point cloud features. The method involves using genetic algorithms to iteratively evolve and refine filter parameter settings for radius and neighbor point threshold based on point cloud characteristics. This avoids manual parameter tuning and improves noise recall rate compared to traditional methods.

Source

A method for processing lidar radar echo data that improves fuzzy neural network filtering to extract useful signals from noisy lidar data. The method uses a fuzzy adaptive filter with 7 rules to eliminate noise interference from lidar signals without requiring transfer functions of the acquisition system. The filter adjusts variables using an improved LMS algorithm to converge precisely. This allows limiting noise to arbitrary precision, improving lidar signal processing compared to traditional methods.

Source

Calibrating lidar background noise to improve the quality of lidar data used for autonomous vehicle perception. The technique involves comparing lidar return signals during the pulse round trip time with signals during a dwell time between pulses. By analyzing the differences, it determines the lidar background noise level in the environment. This calibrated background noise is then used to adjust the lidar return data to improve object detection accuracy. The technique allows more accurate object detection and trajectory planning by compensating for environmental factors like solar radiation that can obscure lidar returns.

Source

Full waveform lidar data denoising using a convolutional neural network (CNN) for improving the accuracy of depth measurement in lidar systems by removing noise from full waveform lidar data. The method involves using a specially designed CNN architecture with dense blocks and transition layers to denoise lidar data without relying on manual selection of high-frequency components like in empirical mode decomposition. The CNN architecture has 8 dense blocks and transition layers with 2-step convolutions to learn high-dimensional features for denoising.

Source

Lidar system that uses vortex light fields to improve daytime signal accuracy by filtering out background light and noise. The system generates a vortex light field from a Gaussian light source, emits it into the atmosphere, and receives the backscattered signals. By focusing and separating the vortex light signals, it extracts the useful echoes from the background light and noise. This allows better inversion of lidar data in daytime conditions where atmospheric light can degrade accuracy.

Source

Real-time estimation of DC bias and noise power for light detection and ranging (LiDAR) systems to improve accuracy and efficiency of LiDAR by extracting the returned laser beam signal more accurately and efficiently in real time. The method involves determining estimated DC bias and noise power using noise data from consecutive time windows. The estimated bias is calculated for the first window using that window's noise, and then the estimated bias is aggregated with the instantaneous bias from the second window to get an improved estimated bias for the second window. This iterative process allows optimizing the amount of noise data used in the calculation to improve calculation speed. The estimated bias and noise power can then be used to calculate the detection threshold for LiDAR signal extraction.

Source

Laser 3D imaging through clouds and fog using photon counting entropy to enhance target extraction. The method involves distinguishing target pixels from non-target pixels based on photon counting, estimating cloud backscatter from non-target pixels, subtracting it from target pixels, and using photon counting sliding windows to filter remaining noise. This improves signal-to-noise ratio and enables precise target range imaging in cloudy environments.

Source

A method for denoising lidar signals used in atmospheric particle monitoring to reduce background noise and extract effective signals. The method involves fitting the lidar signal points with a piecewise cubic spline function and filtering noise through the spline function. The spline coefficients are controlled by a cost function to optimize the fit. The cost function takes a minimum value when the required spline is found.

Source

Lidar signal processing method to improve measurement accuracy by dynamically adjusting gain based on echo signal characteristics. The method involves calculating a target gain based on the number of signal points above a threshold in the first echo signal. This adjusted gain is then applied to the next echo signal to recover weak or saturated signals. By tailoring gain to specific echo levels, it prevents saturation or loss of weak signals. The gain adjustment is calculated from the echo signal itself rather than fixed values, adapting to different signal dynamics.

Source

A method for processing single-photon lidar data to denoise and enhance point clouds from complex scenes with low signal-to-noise ratios. The method uses both temporal correlation and spatial correlation to denoise and fill missing points. It involves adaptive windowing based on depth to separate signals and noise, followed by improved K-means clustering using spatial correlation to identify and filter out noise points.

Source

Method for improving lidar detection performance by extracting both radar system noise and environmental noise during signal processing. This involves a technique to effectively extract environmental noise while extracting radar system noise. It enables better noise processing and radar performance by accurately modeling and removing both sources of noise instead of just one.

Source

Signal processing method for lidar noise reduction that improves noise detection rate by extracting features from multiple echoes to identify and remove noise echos. The method involves detecting echoes in the lidar signal, extracting amplitude at each moment, and determining if there are echoes. If so, feature detection is performed on non-last echos to identify noise echos. These noise echos are removed from the signal to get a filtered reference reflection. Target echoes are then identified and packaged from the filtered reference signal. This filtered signal with target echoes is further processed to generate point cloud data. By isolating and removing noise echos based on their features, it reduces false detections.

Source

Using a bidirectional gated recurrent neural network (BiGRU) to restore lidar signals with saturation issues. The BiGRU network takes short segments of saturated lidar signals as input and outputs restored signals. The BiGRU architecture allows the network to learn the dynamics of lidar signals and recover accurate range and intensity values even from saturated inputs.

Source

Real-time anti-interference method for lidar to improve its performance in interference environments without sacrificing real-time detection capability. The method involves sending multiple laser pulses with inconsistent timings and matching the echoes in received signals to identify the stable target echoes. This allows filtering out interference echoes that are unstable. The filtered echoes are then further processed to correct and restore the pulse width. By detecting multiple times with varying delays, the fixed target echoes can be isolated from unstable interference echoes.

Source

Efficiently collecting and processing lidar noise data to improve autonomous vehicle perception in challenging weather conditions. The method involves using millimeter wave radar to identify false targets in lidar data. By comparing lidar points to millimeter wave radar targets, noise false targets in lidar can be identified. This allows selective retraining of the lidar target detection algorithm using the collected false target points to improve robustness. The method enables accurate and efficient mining of lidar noise false targets caused by rain, snow, fog, and dust.

Source

Denosing lidar signals to improve accuracy and range of lidar systems. The method involves preprocessing the lidar signals by steps of background signal deduction and geometric overlap factor correction. This is followed by signal denoising using the Empirical Mode Decomposition (EMD) algorithm to decompose the signal into intrinsic mode functions representing different scales. The lower scale intrinsic mode functions representing noise are removed to denoise the signal.

Source

Denosing photon counting lidar point clouds in complex scenes using a multi-peak Gaussian fitting algorithm. The algorithm fits a Gaussian distribution to the multi-modal density histogram of the point cloud to adaptively calculate noise and signal thresholds for precise denoising. The basic idea is that signal points are denser than noise points, even in complex environments with uneven density. The Gaussian fitting allows adaptive threshold calculation based on the density distribution.

Source

Lidar noise reduction technique for autonomous vehicles that minimizes noise in lidar point clouds for improved accuracy. The technique involves a multi-modal noise filtering process applied at both grid and point levels to optimize noise removal. It generates lidar free space data by first selecting lidar points within the vehicle's region of interest. Then it processes the points using a 2D grid map to filter grid-level noise. Finally, it filters point-level noise. This multi-modal approach balances accuracy and efficiency for real-time noise reduction of lidar input data.

Source

Noise reduction method for laser radar point cloud intensity images that robustly removes multiplicative speckle noise. The method involves converting irregular point cloud data into regular gridded images, then decomposing the gridded images using wavelet transforms and reducing noise in the high-frequency components. The low-frequency components are further denoised using non-local mean filtering. This approach converts multiplicative speckle noise into additive noise that is easier to separate and remove using wavelet transforms and thresholding.

Source

Method to improve the detection range of wind-measuring lidar by combining variational mode decomposition (VMD) and hierarchical bottom-up agglomerative clustering (HBA) for denoising lidar signals. The method involves: 1) Decomposing the lidar signal using VMD with optimized parameters found using HBA clustering to extract the useful scattered signal from noise. 2) Fusing the decomposed signals using HBA to reconstruct the denoised lidar signal with improved detection range compared to VMD alone.

Source

Echo signal processing for LiDAR systems with improved ranging accuracy. The method involves converting an analog LiDAR echo signal into digital pulses using multiple thresholds. This allows more efficient and accurate processing compared to using a single threshold. The echo signal is compared against multiple thresholds to generate multiple digital signals representing different portions of the echo. This provides more detailed information about the echo characteristics. The digital signals are then analyzed to determine if there are issues like echo superposition or noise. If so, additional processing like noise reduction is performed to improve ranging accuracy.

Source

A technique to reduce background light interference in LiDAR. The technique uses an opaque material with a small aperture behind the lens. Light is focused through the aperture onto a waveguide. A mirror reflects the light towards an array of detectors. This selectively filters out background light not aimed through the aperture.

Source

Coaxial LiDAR system for improved light collection efficiency to improve signal-to-noise ratio and range of LiDAR detection. The coaxial LiDAR system uses a non-reciprocal polarization rotator (like a Faraday rotator) to convert the polarization of the outgoing scanning beam so it is orthogonal to the polarization of the reflected return beam. This allows a polarization beam splitter to separate the two beams and direct the return beam to the photodetector.

Source

Reducing noise in lidar point cloud data by accurately detecting and filtering out raindrop reflections from the front field of view. It involves using polarizers and analyzers in the lidar optics to convert the laser output into linearly polarized light and filter it. This reduces noise from raindrops in the front field of view. For the rear view, a distance filter removes points further than a threshold from adjacent points. This targeted filtering improves noise removal compared to global outlier detection.

Source

FMCW LiDAR system that uses a coherent receiver in the reference optical path to improve target detection. The coherent receiver includes a 90° optical hybrid that extracts the full complex beat signal. Combining the outputs of the optical hybrid suppresses the negative image of the beat frequency. This improves the linear phase noise estimation of the optical source and boosts the target signal-to-noise ratio.

Source

Fusing lidar and millimeter-wave radar data to improve sensor performance for autonomous vehicles in bad weather. The method involves using both lidar and millimeter-wave radar to track targets simultaneously. The millimeter-wave radar provides supplemental data to fuse with the lidar data, improving ranging accuracy and reducing noise compared to just using lidar in bad weather. The fusion is done by extracting features from the lidar waveform like slope rates, pulse widths, and Gaussian curve residuals to classify signals as pure targets or noisy echoes. This allows selecting the lidar samples containing targets for fusion instead of all lidar data. The millimeter-wave radar provides accurate ranging regardless of weather.

Source

Removing noise from LiDAR sensor data to improve precision and robustness of autonomous driving applications. The method involves filtering lidar points using time and intensity information. It separates ground points from the point cloud to reduce calculation, adds intensity and time weights, sets adjustable radius and neighbor count, calculates neighbors, and removes points with low neighbor count.

Source

A method for improving lidar (laser radar) signals to enhance accuracy and reliability by denoising and filtering the lidar signals. The method involves using variational calculus to decompose the lidar signal into multiple components with fixed frequencies and bandwidths. This variational problem is constrained to have the sum of all components equal to the original signal. By finding the optimal decomposition, it allows separating the signal from noise and interference while preserving the true lidar signal components.

Source

A LiDAR device that uses a collimator to generate parallel laser beams from non-parallel beams reflected off a rotating scanner. The collimator actively synchronizes and changes its properties as the scanner rotates to properly collimate the non-parallel beams into parallel beams. This allows using a smaller and faster rotating scanner while still achieving high spatial resolution and signal-to-noise ratio when scanning smaller objects. The active synchronization involves changing the refractive index of the collimator using phase change materials and electrodes.

Source

A LiDAR system that can reduce noise-light interference from sunlight when using a wavelength-tunable optical phased array (OPA) light source. The system uses an active device to adjust the filter to match the OPA's current wavelength, rather than using a fixed band-pass filter. This allows tuning the OPA without increasing noise-light.

Source

Computer unit for lidar devices that can filter out excessive radiation effects from laser signals to improve object detection accuracy in challenging environments. The computer unit processes lidar data points to identify and correct areas of crosstalk and over-radiation that can occur around retroreflective objects. It determines which data points can be assigned to retroreflectors based on analyzing the reflected laser intensity levels. By explicitly identifying areas of excessive radiation and preserving true object data points, it mitigates distortion effects like blooming around retroreflectors that can impair lidar perception.

Source

Steering a light pulse along an optical path and detecting the scattered light to determine a distance to the object. The system uses an array detector where only a subset of detector segments are activated based on the pulse steering direction. This allows steering consecutive pulses to different locations while binocularly collecting the scattered light using a smaller detector subset. Selecting the detector segments that receive the light scattered from the steering direction, can optimize the detection efficiency and reduce background noise compared to using a full detector array.

Source

LiDAR apparatus and method that accurately detects the flight time of laser pulses for ranging in noisy environments. It converts received signals into unipolar signals and analyzes their correlation with reference signals to identify the exact flight time point. This enables robust ranging even when signals are weak or corrupted by noise. If correlation peaks are not detectable, it increases the reference signal intensity or averages multiple measurements to enhance the signals over the noise.

Source

Frequency Modulated Continuous Wave (FMCW) LiDAR sensor that eliminates noise caused by a DC offset between the reference and return laser signals. The FMCW LiDAR system uses two lasers - one for the ranging signal and one for the local oscillator signal. The local oscillator signal is offset from the ranging signal by a predetermined frequency. This non-zero offset eliminates the DC component when the reference and return signals are mixed, reducing noise caused by self-mixing and other factors.

Source

Reducing false detections and improving accuracy in FMCW LiDAR systems by analyzing noise characteristics in each frequency band and applying different filtering thresholds. The method involves dividing the frequency bands for up and down chirps, analyzing the noise shape and size in each band, setting threshold coefficients based on the analysis, calculating band-specific thresholds using the coefficients, and filtering noise in each band using the calculated thresholds to detect valid data.

Source

Removing noise from LIDAR data caused by sunlight to improve object recognition. The noise removal apparatus predicts the direction and position of the sun based on GPS and image data. It then identifies a region of interest in the LIDAR data corresponding to the sun's location and removes noise points in that region. By selectively filtering out sunlight-induced noise, the apparatus aims to improve detection accuracy without losing actual object data.

Source

Scanning LIDAR systems that avoid errors caused by external light sources like ambient light or other LIDAR systems. It uses a modulated laser pulse that is distinguishable from ambient light and other LIDAR systems. The LIDAR system includes a pulsed laser that emits a modulated optical pulse. The modulation can be frequency, amplitude, phase or code modulation. The LIDAR system also includes a receiver to detect the modulated laser pulse reflections. The receiver filters out non-modulated light sources, like ambient light or other LIDAR systems, using the modulation frequency or code.

Source

A method and system for processing lidar point cloud data using deep learning neural networks instead of traditional filtering techniques. The method involves receiving lidar point cloud data, determining if preprocessing is needed based on resolution and signal strength, and then using a pretrained depth convolutional neural network to enhance, denoise, or amplify the point cloud as needed. This allows customized preprocessing for different conditions rather than generic filtering.

Source

Filtering method for improving the quality of low-quality lidar one-dimensional range images with poor signal-to-noise ratios. The filter combines Savitzky-Golay filtering for smoothing and L0 gradient minimization for preserving high-frequency details. It involves sliding window denoising of the time-distance range data using a polynomial fitting and least squares estimation at each window center. This joint filtering provides stable, low-jitter images with improved quality compared to traditional denoising methods for lidar range data.

Source

Reducing noise in lidar systems caused by laser reflection from an internal absorption member. The lidar system has a rotating mirror that changes the laser direction between reflections from a reflector and an absorber. By comparing lidar return signals with and without absorber reflection, the system can estimate and subtract the absorber reflection noise from the lidar output signal.

Source

A LiDAR system with descan compensation to mitigate signal losses due to descan in fast scanning LiDAR systems. The technique involves intentionally decentering the local oscillator (LO) signal from the optical axis on the second lens to increase the overlap with the target return signal at the detector. This offsets the LO signal to compensate for spatial misalignment caused by fast scanning mirror speeds that can reduce mixing efficiency.

Source

A LIDAR system that uses cross-polarized light to determine target material and orientation, and reduce speckle noise. The system sends out a co-propagating, cross-polarized beam towards the target. The detectors measure the returned signals. By comparing the signal strengths from the detectors, the system can determine target properties like reflectivity and orientation. This is possible because different materials reflect polarized light differently. The cross-polarized beam also mitigates speckle noise compared to traditional LIDAR beams.

Source

Removing noise from triangular ranging lidar data to improve accuracy in environments with strong light interference. The method involves detecting and removing noise points from the triangular ranging measurements using a straight line algorithm. The algorithm identifies continuous sets of N sampling points that lie on a straight line pointing back towards the lidar's origin. These sets of points are flagged as noise and removed. This helps mitigate interference from bright spots that can be mistaken as laser reflections by the lidar's camera.

Source

Method for denoising and filtering spaceborne photon counting lidar data to reduce noise in lidar data collected using micropulse lasers. The method involves using machine learning models to identify and filter out noise photons in the lidar data. It segments the lidar data into subsets, extracts photon features, trains a stacked model with sub-models like logistic regression and random forests, and uses it to classify lidar points as signal or noise. The stacked model combines the sub-models to improve noise filtering. The method allows fast denoising of lidar data compared to other methods for spaceborne photon counting lidar.

Source

Ranging operation in LiDAR systems that leverages the dynamic range of the receiver to enable accurate distance and reflectivity measurements across a wide range of target distances and reflectivities. The method involves transmitting two signals with carefully chosen levels and a time gap. This allows at least one of the reflected signals to fall within the linear dynamic range of the receiver. The received signals are identified and compared to determine distance and reflectivity.

Source

A method of filtering LIDAR point clouds to reduce noise while preserving reflected light pulses. It involves separately filtering pulses based on length and strength. Longer pulses have a lower threshold since noise is less common, while shorter pulses have a higher threshold to avoid filtering out reflected pulses. This customized thresholding improves noise reduction without mistakenly removing actual signals. The thresholds can adapt based on background light levels.

Source

Improving the range accuracy of a lidar system for autonomous vehicles by using a hybrid digitization technique. The lidar system sends pulses of light towards objects, then receives and digitizes the reflected pulses. To optimize range accuracy across short and long distances, it samples the received analog sensor data at a lower rate for weak pulses, and at a higher rate for stronger pulses. This extracts timing data more accurately from the weak pulses, while still using the full amplitude data from the stronger pulses.

Source

Method and apparatus for removing noise from reflective objects in 3D scans by fusing data from multiple sensors. The method involves determining a reflection region using point cloud and intensity data from a primary sensor. Noise in other sensors is then found by comparing points in the reflection region. This noise is removed from the secondary sensor data and the primary scan is merged with the denoised secondary scans.

Source

Ambient light denoising method for lidar systems that improves lidar performance in various lighting conditions by adaptively setting noise thresholds based on scene-specific ambient light levels. The method involves monitoring the number of optical excitation signals generated by a photodetector in response to laser pulses reflecting off objects in a scene. By counting the signals over a period of time, the total excitation signal level is determined for each pixel. These levels are used to adaptively set noise thresholds for ambient light denoising to account for variations in ambient light intensity across different scenes and angles.

Source

A lidar imaging method and system using a Gm-APD array sensor to improve image quality under strong background noise conditions. The method involves acquiring two sets of cumulative lidar detection data with the target in and out of the lidar's range gate. Comparing the statistical histograms of these sets allows determination of the target's range interval. Then, an imaging algorithm is applied only within this interval to suppress noise from other ranges. This improves range information recovery and target-background contrast in lidar images.

Source