3D Sensor Object Identification With Fewer LiDAR Scan Lines

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing LiDAR systems face challenges in achieving high precision object identification with a small number of scan lines, leading to increased calculation costs and limitations in low-cost, low-end processing devices.

Innovation Solution

An object identification system using a three-dimensional sensor generating multiple items of line data for different heights, combined with a processing device that includes first and second neural networks to integrate intermediate data, reducing height-direction dependence and allowing object classification with a small number of horizontal lines.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the resolution of point group data is increased to improve object identification precision, then measurement precision is improved, but calculation costs increase drastically

Engineering Contradiction:
Improveobject identification precisionVSAvoidcalculation costs
Core Design Contradiction:
Measurement precisionVSPower

Solution Approach 1:

The patent segments the LiDAR scan data into multiple horizontal lines at different heights and processes each line independently through first neural networks to generate intermediate data. This segmentation allows the system to handle data in manageable portions, reducing overall calculation complexity while maintaining identification precision through the combined results of processed segments.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts and processes only the essential information from each horizontal line through the first neural networks, generating intermediate data that captures key object characteristics. This extraction approach filters out redundant information early in the processing pipeline, reducing the data volume that requires intensive computation in subsequent integration stages.

Inventive Principle:
Principle #2Taking out (Extraction)

2Device complexity

If the number of scan lines is reduced to lower processing requirements, then device complexity is reduced, but object identification precision deteriorates

Engineering Contradiction:
Improveprocessing device requirementsVSAvoidobject identification precision
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent divides the scan data into multiple horizontal lines processed by separate first neural networks, then combines their intermediate outputs through a second neural network. This segmentation strategy enables the system to achieve high identification precision with fewer scan lines by efficiently processing and integrating information from each line independently, reducing overall processing requirements.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial processing by using first neural networks to generate intermediate data from each horizontal line, then selectively integrating this intermediate data through a second neural network. This partial action approach processes only the necessary information from each scan line, achieving sufficient precision without requiring complete processing of all data points, thus reducing device complexity.

Inventive Principle:
Principle #16Partial or excessive action

3Productivity

If a small number of horizontal lines are used to reduce calculation costs, then productivity is improved, but measurement precision decreases

Engineering Contradiction:
Improveprocessing efficiencyVSAvoidobject identification precision
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent performs preliminary processing by using first neural networks to generate intermediate data from each horizontal line before final integration. This preliminary action prepares and pre-processes the data in an efficient manner, enabling the system to achieve high productivity with fewer scan lines while maintaining precision through the structured integration of pre-processed intermediate data.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces intermediate data as a mediator between the raw horizontal line data and the final object identification results. The first neural networks generate this intermediate representation, which then serves as input to the second neural network for final classification. This intermediary approach enables efficient processing with fewer scan lines by creating a compressed, information-rich representation that maintains precision while improving productivity.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3767338B1Object identification system, automobile, vehicular lamp fitting, and method for identifying type of object
Publication Date: 2026.01.28 KOITO MFG CO LTD
  • EP3767338B1 patent drawingFigure 1
  • EP3767338B1 patent drawingFigure 2
  • EP3767338B1 patent drawingFigure 3A~3D

AI summary

A three-dimensional sensor 20 generates multiple items of line data LD1 through LDN with respect to multiple horizontal lines L1 through LN arranged at different heights. Multiple first neural networks 72 each generate first intermediate data MD relating to a corresponding item from among the multiple items of line data LD. Each first intermediate data MD1 indicates the probability of matching between the corresponding line data and each of multiple portions of multiple kinds. A combining processing unit 74 receives the multiple items of first intermediate data MD1, and combines the first intermediate data thus received so as to generate at least one item of second intermediate data MD2. A second neural network 76 receives the at least one item of second intermediate data MD2, and generates final data FD that indicates the probability of matching between the object OBJ and each of the multiple kinds.