Attention-Driven Object Detection for Low-Compute Autonomous Driving

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Autonomous driving technologies require significant computing power, leading to performance drops and potential safety risks due to reduced computing resources.

Innovation Solution

A method utilizing attention-driven resource allocation through a reinforcement learning algorithm and confidence scores to enhance object detection accuracy while minimizing computing power consumption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If computing power is reduced to save energy, then energy consumption decreases, but autonomous driving performance drops and safety is compromised

Engineering Contradiction:
Improvecomputing power consumptionVSAvoidautonomous driving safety
Core Design Contradiction:
Use of energy by moving objectVSReliability

Solution Approach 1:

The patent segments the object detection process into two phases: an initial detection phase that processes the entire image, and a re-detection phase that selectively processes only unconfident regions identified by low confidence scores. This segmentation allows the system to reduce computing power consumption while maintaining detection accuracy in critical areas.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by directing enhanced computational resources (re-detection) only to specific regions with low confidence scores rather than uniformly processing the entire image. This ensures high reliability in uncertain regions while conserving energy in regions where the initial detection is already confident.

Inventive Principle:
Principle #3Local quality

2Use of energy by moving object

If computing power is reduced, then energy consumption decreases, but object detection accuracy drops

Engineering Contradiction:
Improvecomputing power consumptionVSAvoidobject detection accuracy
Core Design Contradiction:
Use of energy by moving objectVSMeasurement precision

Solution Approach 1:

The patent applies partial action by performing object detection only on specific unconfident regions rather than the entire image. The reinforcement learning agent selectively identifies regions requiring re-detection based on confidence scores, applying computational resources partially where needed rather than excessively across the whole image, thus reducing energy consumption while maintaining accuracy.

Inventive Principle:
Principle #16Partial or excessive action

3Use of energy by moving object

If reinforcement learning is used to select regions for re-detection, then computing power is saved, but system complexity increases

Engineering Contradiction:
Improvecomputing power consumptionVSAvoidsystem complexity
Core Design Contradiction:
Use of energy by moving objectVSDevice complexity

Solution Approach 1:

The patent implements feedback through confidence scores that guide the reinforcement learning agent's region selection. The agent uses the confidence information from the initial detection as feedback to identify unconfident regions, creating a closed-loop system that adaptively allocates computing resources while managing complexity through information-driven decision making.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3690719B1Method and device for attention-driven resource allocation by using AVM and reinforcement learning to thereby achieve safety of autonoumous driving
Publication Date: 2025.11.19 STRADVISION
  • EP3690719B1 patent drawingFigure 1
  • EP3690719B1 patent drawingFigure 2
  • EP3690719B1 patent drawingFigure 3

AI summary

A method for achieving better performance in an autonomous driving while saving computing powers, by using confidence scores representing a credibility of an object detection which is generated in parallel with an object detection process is provided. And the method includes steps of: (a) a computing device acquiring at least one circumstance image on surroundings of a subject vehicle, through at least one panorama view sensor installed on the subject vehicle; (b) the computing device instructing a Convolutional Neural Network(CNN) to apply at least one CNN operation to the circumstance image, to thereby generate initial object information and initial confidence information on the circumstance image; and (c) the computing device generating final object information on the circumstance image by referring to the initial object information and the initial confidence information, with a support of an RL agent.