LiDAR Target Classification Using Photon-Histogram Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Lidar systems face high computational effort in optical distance measurements, leading to false detections due to inability to distinguish between object reflections and environmental conditions like rain or fog, posing safety risks in autonomous vehicles.
Innovation Solution
Employing a neuronal network or Gaussian Mixture Model (GMM) to analyze photon histograms from Lidar sensors, utilizing fully connected neuronal networks (FCN) and convolutional neuronal networks (CNN) to detect patterns and classify targets as objects or environmental conditions, reducing computational complexity by focusing on peak features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If simplified models and filter approaches are used to reduce computational effort, then processing speed is improved, but the ability to distinguish between object reflections and environmental conditions deteriorates
Solution Approach 1:
The patent replaces traditional mechanical filter approaches with a neural network-based system that processes photon histogram data. The neural network learns to distinguish between object reflections and environmental conditions through training, achieving both high processing speed and high detection accuracy without the trade-off present in traditional filtering methods.
2Measurement precision
If 3D calculations are performed to improve object detection accuracy, then detection precision is improved, but computational complexity increases
Solution Approach 1:
The patent extracts only the essential features from the raw Lidar data by analyzing photon histograms and identifying characteristic patterns. Instead of performing full 3D calculations on all data points, the system extracts relevant temporal and intensity features from photon return data, significantly reducing computational complexity while maintaining detection accuracy.
Solution Approach 2:
The patent transforms the problem from spatial 3D coordinate processing to temporal histogram analysis. By changing the parameter space from (x, y, z) coordinates to time-of-flight distributed photon counts, the system achieves accurate object detection with reduced computational complexity through the neural network's ability to process histogram patterns efficiently.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enhances target detection and classification in Lidar systems, enabling precise semantic segmentation and reducing computational effort, allowing robust differentiation between object and environmental reflections, thus improving safety in autonomous driving.
Implementation Method 1
These are based upon the so-called time-of-flight principle. In the process, a scanning sensor is used, which periodically emits pulses. The pulses are reflected by targets, wherein the reflected pulses are detected. By determining the transit time of the pulses from the sensor to the target and back again, the distance to this target can be inferred with the help of the speed of light.
Data Source
AI summary
A method for classifying targets is proposed, which comprises the extraction of features from measurement data of one or several receiving elements of a sensor by means of a neuronal network or by means of a Gaussian Mixture Model, wherein the respective measurement data of the at least one receiving element of the sensor involve at least one section of a photon histogram, and wherein the neuronal network involves a fully connected neuronal network or a convolutional neuronal network.