Road Obstacle Detection via Semantic Label Reconstruction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing road obstacle detection systems face challenges in accurately identifying obstacles due to errors in semantic labeling, leading to reduced detection accuracy.

Innovation Solution

A road obstacle detection device that estimates semantic labels for each pixel of an image using a learned classifier, generates a semantic label image, and reconstructs the original image for comparison with the detected image to identify obstacles, thereby improving detection accuracy by highlighting deviations from the original image.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If a semantic label is assigned to an obstacle as "an obstacle or other" using a classifier, then the detection process is simplified, but the detection accuracy deteriorates due to label assignment failures

Engineering Contradiction:

Improvedetection process simplicityVSAvoiddetection accuracy

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The detection process is segmented into two independent stages: first generating a semantic label image using a classifier, then reconstructing the original image from semantic labels and comparing with the actual image. This segmentation allows each stage to optimize for its specific function while maintaining overall system simplicity.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

An image reconstruction unit acts as an intermediary between the semantic label generation and obstacle detection. It reconstructs the original image from semantic labels and compares it with the actual captured image, thereby identifying obstacles without directly relying on potentially inaccurate semantic label assignments.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If the original image is reconstructed from the semantic label image and compared with the detected image, then the detection accuracy is improved, but the device complexity increases

Engineering Contradiction:

Improvedetection accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

Instead of directly processing the original image for obstacle detection, the system creates a copy (reconstructed image) from the semantic label image. This copied image serves as a reference for comparison, enabling accurate obstacle detection while keeping the processing of each image independent and manageable.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system performs preliminary reconstruction of the original image from semantic labels before the actual obstacle detection comparison. This preliminary action prepares both images (original and reconstructed) in advance, making the subsequent detection process more efficient and accurate without adding complexity during the critical detection phase.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11704828B2Road obstacle detection device, road obstacle detection method, and computer-readable storage medium

Publication Date: 2023.07.18 TOYOTA JIDOSHA KK

AI summary

The road obstacle detection device includes a semantic label estimation unit that estimates a semantic label for each pixel of an image using a classifier learned in advance and generates a semantic label image, an original image estimation unit for reconstruction of the original image from the semantic label image, a difference calculating unit for calculating a difference between the original image and the reconstructed image from the original image estimation unit as a calculation result, and a road obstacle detection unit for detecting a road obstacle based on the calculation result.