Road Image ROI Extraction for Efficient Long-Distance Object Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing object detection methods in autonomous driving and robot-following scenarios suffer from low accuracy and efficiency.

Innovation Solution

A method and apparatus that extracts a region of interest (ROI) from an original image, focusing on the pixel point farthest from the sensor, and performs object detection within this ROI, utilizing road segmentation and depth information to enhance accuracy and efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If object detection is performed on the entire original image, then comprehensive detection coverage is achieved, but computing resources are wasted and efficiency is reduced

Engineering Contradiction:
Improvedetection efficiencyVSAvoidcomputing resource consumption
Core Design Contradiction:
ProductivityVSLoss of energy

Solution Approach 1:

The patent divides the original image into multiple regions of interest (ROIs) based on road segmentation results and depth information. Instead of processing the entire image, the system segments it into relevant areas where objects are likely to be detected, thereby reducing computational load while maintaining detection effectiveness.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts specific regions of interest from the original image based on depth information and road segmentation. By taking out only the relevant portions of the image where objects are likely to exist, the system avoids processing unnecessary areas, thus improving efficiency and reducing energy consumption.

Inventive Principle:
Principle #2Taking out (Extraction)

2Measurement precision

If object detection is performed on the entire original image, then all potential objects are detected, but detection accuracy for distant objects is reduced

Engineering Contradiction:
Improvedetection accuracyVSAvoiddetection area
Core Design Contradiction:
Measurement precisionVSArea of stationary object

Solution Approach 1:

The patent applies different processing qualities to different regions of the image based on depth information. Regions with greater depth (distant areas) are processed with higher quality and more focused attention, while closer regions use standard processing. This local quality adjustment improves detection accuracy for distant objects without unnecessarily processing the entire image at high quality.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent introduces depth information as an additional dimension for region selection. By using depth maps and depth-based filtering, the system identifies and focuses on distant regions that require higher detection accuracy, adding a depth dimension to the traditional 2D image processing approach.

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

3Reliability

If the ROI extraction includes all pixel points exceeding a preset distance, then comprehensive long-distance detection is achieved, but processing complexity increases

Engineering Contradiction:
Improvelong-distance detection reliabilityVSAvoidprocessing complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent performs preliminary road segmentation and depth-based filtering before object detection. By pre-identifying regions that exceed the preset distance threshold and belong to road surfaces, the system prepares the data in advance, which simplifies the subsequent object detection process and reduces overall processing complexity while maintaining reliable long-distance detection.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4246455B1Method and device for detecting object and vehicle
Publication Date: 2025.09.17 XIAOMI EV TECH CO LTD
  • EP4246455B1 patent drawingFigure 1~2
  • EP4246455B1 patent drawingFigure 3~5
  • EP4246455B1 patent drawingFigure 6~8

AI summary

A method for detecting an object includes obtaining (S101) an original image of a road through a sensor by performing an image capturing operation; extracting (S102) a region of interest (ROI) including a first target pixel point from the original image, wherein the first target pixel point is a pixel point in the original image corresponding to a position on the road that is farthest from the sensor; and generating (SI03) an object detection result by performing object detection on the ROI.