Multi-Range Object Detection With Multi-Resolution Grid Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing grid-based perception systems in automotive perception systems face challenges with quadratic increases in memory requirements and computational costs due to varying spatial ranges and resolutions, which are not efficiently addressed by current data representation methods.
Innovation Solution
A unified neural network structure processes sensor-based data using multiple branches with different spatial resolutions, interacting at intermediate stages to fuse outputs and efficiently manage memory and computation demands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single high-resolution Cartesian grid is used to cover a large spatial range, then detection precision is improved, but memory requirements and computational costs increase quadratically
Solution Approach 1:
The patent divides the large spatial range into multiple Cartesian grids with different spatial resolutions. A first Cartesian grid covers a first spatial range with a first spatial resolution, while a second Cartesian grid covers a second spatial range with a second spatial resolution. This segmentation allows each grid to use appropriate resolution for its range, avoiding the quadratic memory increase that would result from using a single high-resolution grid for the entire large spatial range.
2Quantity of substance
If multiple Cartesian grids with different spatial resolutions are used, then memory efficiency is improved, but feature alignment and integration complexity increases
Solution Approach 1:
The patent employs parameter changes by dynamically adjusting the spatial resolution parameter for different Cartesian grids based on their spatial ranges. The first Cartesian grid uses a first spatial resolution appropriate for its first spatial range, while the second Cartesian grid uses a second spatial resolution for its second spatial range. This parameter adaptation allows efficient memory usage while the neural network learns to integrate features across these different resolutions through trained transformation layers.
Data Source
AI summary
Provided is a method for object detection in a surrounding of a vehicle using a deep neural network, comprising: inputting a first set of sensor-based data for a first Cartesian grid having a first spatial dimension and a first spatial resolution into a first branch of the deep neural network; inputting a second set of sensor-based data for a second Cartesian grid having a second spatial dimension and a second spatial resolution into a second branch of the deep neural network; providing an interaction between the first branch of the deep neural network and the second branch of the deep neural network at an intermediate stage of the deep neural network; and fusing a first output of the first branch of the deep neural network and a second output of the second branch of the deep neural network to detect the object in the surrounding of the vehicle.