Adaptive Image Resolution for Faster Vehicle Object Detection

Solution Overview

Problem

Existing object detection methods in vehicles, such as those used in ADAS and autonomous driving, face challenges in accurately and efficiently detecting objects at both close and far distances, with CNN-based models requiring significant resources and filter-based methods being sensitive to object posture transformations and occlusions.

Innovation Solution

A method involving downscaling original images to multiple resolutions, using a neural network-based object detection model, and determining a target image based on previous object detection counts to optimize resource usage and processing time.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If CNN-based object detection models are used to detect objects in images, then object detection accuracy is improved, but resource consumption and processing time increase significantly

Engineering Contradiction:

Improveobject detection accuracyVSAvoidprocessing speed

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The image is divided into multiple regions of interest (ROIs) based on detected interest points. Only these specific regions are processed by the CNN model instead of the entire image, reducing computational load while maintaining detection accuracy for critical objects.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Interest points are detected and ROIs are identified before applying the CNN model. This preliminary processing step prepares the data in advance, allowing the CNN to focus only on relevant regions and improving overall processing efficiency.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If CNN-based object detection models are used to detect objects in images, then object detection accuracy is improved, but computational resources required increase

Engineering Contradiction:

Improveobject detection accuracyVSAvoidcomputational resource consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The image is divided into multiple regions of interest (ROIs) based on detected interest points. Only these specific regions are processed by the CNN model instead of the entire image, reducing computational load while maintaining detection accuracy for critical objects.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Instead of processing the entire image with the computationally intensive CNN model, only partial regions (ROIs containing interest points) are processed. This partial action approach reduces resource consumption while still achieving accurate detection where needed.

Inventive Principle:

Principle #16Partial or excessive action

3Productivity

If filter-based object detection methods are used, then processing speed is improved, but detection accuracy deteriorates due to sensitivity to object posture transformations and occlusions

Engineering Contradiction:

Improveprocessing speedVSAvoidobject detection accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

Interest points are detected and ROIs are identified before applying the CNN model. This preliminary processing step prepares the data in advance, allowing the CNN to focus only on relevant regions and improving overall processing efficiency.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary step between simple filtering and full CNN processing: interest point detection and ROI extraction. This intermediary approach combines the speed of filter-based methods with the accuracy of CNNs by preparing targeted regions for deeper processing.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS12450689B2Object detection method and device

Publication Date: 2025.10.21 NEXTCHIP

AI summary

Provided is an object detection method including detecting a first object by using an object detection filter in an original image and images of which resolutions are downscaled and detecting a second object by using a neural network-based object detection model of which an input is a target image that is determined to be one of the original image and the images of which resolutions are downscaled.