Multi-Task Neural Network for Vehicle Perception Point Cloud Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Automotive perception systems face challenges in efficiently and accurately obtaining different kinds of outputs, such as object detection and semantic segmentation, due to the cumbersome nature of current methods.

Innovation Solution

A computer-implemented method using a multi-task and multi-resolution neural network architecture that processes point clouds and grid structures through artificial neural networks, enabling hybrid processing and efficient computation for various tasks and resolutions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional separate processing methods are used for different perception tasks, then each task can be processed independently, but the system becomes cumbersome and computationally inefficient

Engineering Contradiction:
Improveprocessing efficiencyVSAvoidsystem complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent combines multiple perception tasks (object detection, semantic segmentation, etc.) into a unified neural network architecture that processes point cloud data through shared feature extraction layers, eliminating the need for separate processing pipelines and reducing overall system complexity while improving efficiency

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The neural network architecture is designed with universal feature extraction capabilities that can serve multiple perception tasks simultaneously through different output heads, allowing a single system to perform various functions without requiring task-specific separate processors

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Productivity

If early stage gridding is applied to point cloud data, then computational structure is provided, but information loss occurs

Engineering Contradiction:
Improvecomputational efficiencyVSAvoidpoint cloud information loss
Core Design Contradiction:
ProductivityVSLoss of information

Solution Approach 1:

The patent applies feature extraction and processing operations to point cloud data before gridding occurs, preparing the data in a way that preserves critical information while still enabling efficient structured processing in subsequent stages

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent maintains point cloud data in its original point-based representation rather than converting to grid structure early, preserving spatial information while using alternative computational approaches that don't rely on traditional gridding

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Measurement precision

If high resolution grid is used for processing, then detailed information is captured, but computational waste occurs in sparse regions

Engineering Contradiction:
Improvespatial resolutionVSAvoidcomputational energy waste
Core Design Contradiction:
Measurement precisionVSLoss of energy

Solution Approach 1:

The patent applies different processing resolutions to different spatial regions based on their characteristics, using high resolution only where needed and lower resolution in sparse regions, optimizing the balance between information capture and computational efficiency

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4531003A1Methods and systems for determining information related to an environment of a vehicle
Publication Date: 2025.04.02 APTIV TECHNOLOGIES AG
  • EP4531003A1 patent drawingFigure 1
  • EP4531003A1 patent drawingFigure 2A
  • EP4531003A1 patent drawingFigure 2B

AI summary

A computer implemented method for determining information related to an environment of a vehicle comprises the following steps carried out by computer hardware components: determining a point cloud based on measurement data, the point cloud comprising a plurality of points; determining respective features for each point of the point cloud; determining a grid structure based on the point cloud; determining a processed grid structure based on processing the grid structure using an artificial neural network; and determining at least one final grid structure based on the point cloud, the respective features for each point of the point cloud, and the processed grid structure.