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