Scene-Dependent Object Queries for Low-Latency 3D Bounding Boxes

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Autonomous vehicles face challenges in accurately drawing 3D bounding boxes for objects in real-time driving environments due to insufficient semantic data from individual cameras and the processing latency of scene-independent object queries.

Innovation Solution

The system utilizes a combination of scene-dependent and scene-independent object queries, enriched by vision transformers with self-attention and cross-attention functions, to generate bounding boxes, reducing processing demands and improving object identification efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If scene-independent object queries are used to generate bounding boxes, then object detection can be performed, but processing latency increases and processing demands rise

Engineering Contradiction:

Improveobject detection accuracyVSAvoidprocessing latency

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary action by using scene-independent object queries to generate candidate bounding boxes and predictions in advance, then refines these predictions using scene-dependent queries. This two-stage approach allows the system to prepare potential object locations beforehand, reducing the computational burden and latency during real-time detection while maintaining accuracy through subsequent refinement.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If scene-independent object queries are used, then bounding boxes can be generated, but processing demands increase

Engineering Contradiction:

Improveobject detection accuracyVSAvoidprocessing demands

Core Design Contradiction:

Measurement precisionVSPower

Solution Approach 1:

The system segments the object detection process into two distinct stages: a first stage using scene-independent queries to generate candidate predictions, and a second stage using scene-dependent queries to refine these predictions. This segmentation divides the computationally intensive task into manageable parts, reducing overall processing demands while maintaining detection accuracy through specialized processing in each stage.

Inventive Principle:

Principle #1Segmentation

3Device complexity

If individual camera data is used, then simple processing is required, but insufficient semantic data results in poor object identification

Engineering Contradiction:

Improveprocessing simplicityVSAvoidsemantic data sufficiency

Core Design Contradiction:

Device complexityVSLoss of information

Solution Approach 1:

The system merges data from multiple cameras by generating feature maps from images captured by different cameras and combining these feature maps to create enriched scene representations. This merging process integrates semantic information from multiple sources, providing sufficient contextual data for accurate object identification while maintaining processing efficiency through unified feature map generation and combination.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS12525027B2Using scene dependent object queries to generate bounding boxes

Publication Date: 2026.01.13 MOTIONAL AD LLC

AI summary

A perception system may be used to generate bounding boxes for objects in a vehicle scene. The perception system may receive images and feature maps corresponding to the received images. The perception system may generate scene dependent object queries. The perception system may use the generate scene dependent object queries to generate one or more bounding boxes for objects in the vehicle scene.