Locating Element Detection Using Surround-View Semantic Fusion

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current visual locating systems for autonomous vehicles face challenges in detection robustness and accuracy, particularly in dynamic scenes with insufficient illumination or complex textures, and require extensive customization and maintenance for artificial identifiers.

Innovation Solution

A method utilizing a circular-scanning stitched image from fisheye cameras to detect naturally occurring elements like parking spaces, lane lines, and arrows, employing a deep neural network for semantic segmentation and matching to improve detection accuracy and robustness, without the need for site modifications.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If visual SLAM with traditional image algorithms is used to extract key points and calculate self-positioning, then the system can operate without field modifications, but detection robustness deteriorates in scenes with insufficient illumination or complex textures

Engineering Contradiction:
Improveno field modification neededVSAvoiddetection robustness
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The patent transforms the detection approach by changing the parameter being detected - instead of detecting natural image features (key points) that are sensitive to illumination and texture, the system detects artificial locating identifiers with fixed, known parameters. This transformation makes detection robust against environmental variations while still requiring no field modifications, as the identifiers are naturally present in the scene.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If artificially customized locating identifiers are detected to achieve accurate 3D location and direction calculation, then detection accuracy is improved, but device complexity and maintenance cost increase due to extensive customization and arrangement requirements

Engineering Contradiction:
Improve3D location accuracyVSAvoididentifier customization complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies universality by using a general-purpose deep learning object detection model that can detect various types of locating identifiers (different shapes, sizes, and patterns) without requiring custom processing for each type. The system universally handles multiple identifier formats through semantic segmentation and template matching, eliminating the need for complex customization and arrangement while maintaining high detection accuracy.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Measurement precision

If large amounts of specific identifiers are customized and arranged in the scene to enable accurate locating, then location precision is improved, but loss of time and maintenance cost increase due to extensive arrangement and maintenance requirements

Engineering Contradiction:
Improvelocating precisionVSAvoidmaintenance time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent enables the system to automatically adapt to different scenes by using deep learning models that can identify and detect locating identifiers without manual configuration or arrangement. The system performs self-service by automatically learning the characteristics of identifiers in the scene and adjusting detection parameters accordingly, eliminating the need for time-consuming manual arrangement and maintenance of specific identifiers.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3855351B1Locating element detection method, apparatus, device and medium
Publication Date: 2023.10.25 APOLLO INTELLIGENT DRIVING (BEIJING) TECHNOLOGY CO LTD
  • EP3855351B1 patent drawingFigure 1~2
  • EP3855351B1 patent drawingFigure 3
  • EP3855351B1 patent drawingFigure 4~5

AI summary

The present disclosure provides a locating element detection method, apparatus, device and medium, and relates to object detection technologies and is applicable for autonomous parking scenarios. The method includes: obtaining a circular-scanning stitched image around a vehicle; detect the circular-scanning stitched image to determine at least one locating element existing on ambient ground of the vehicle and a semantic type to which each pixel on the circular-scanning stitched image belongs; and performing matching and fusion on the at least one locating element based on the semantic type to obtain a locating element detection result.