Object Recognition Device Using Stereo and Monocular Fusion

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing vehicle detection systems using either stereo or monocular information face challenges in achieving precise detection accuracy, leading to difficulties in controlling vehicles effectively, especially when pedestrians emerge from shadows.

Innovation Solution

An object recognition device that utilizes both stereo and monocular information from a stereo camera to determine pedestrian detection reliability, integrating parallax and image information for accurate pedestrian detection and vehicle control, allowing for adaptive warning and control methods based on detection reliability.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If only stereo information or monocular information is used for obstacle detection, then device complexity is reduced, but detection accuracy deteriorates

Engineering Contradiction:
Improvedetection system complexityVSAvoidobstacle detection accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent combines stereo camera information and monocular camera information into a single detection system. The controller integrates depth maps from the stereo camera with image information from the monocular camera to generate comprehensive obstacle detection results, thereby improving detection accuracy while maintaining reasonable system complexity through unified processing.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent uses a composite detection approach by fusing data from two different camera types (stereo and monocular). This composite information processing method leverages the depth perception capability of stereo cameras and the high-resolution imaging capability of monocular cameras to achieve superior detection accuracy compared to using either camera type alone.

Inventive Principle:
Principle #40Composite materials

2Measurement precision

If stereo information is used for close obstacle detection, then detection accuracy is improved, but response time for distant obstacles deteriorates

Engineering Contradiction:
Improveclose obstacle detection accuracyVSAvoiddistant obstacle detection time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies different detection strategies to different spatial zones. For close obstacles, the system prioritizes stereo information for high-precision detection. For distant obstacles, the system utilizes monocular information to ensure timely detection. This localized quality approach optimizes both accuracy for near objects and response time for far objects.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The detection system dynamically adjusts its information fusion strategy based on obstacle distance. The controller evaluates the distance to detected objects and adaptively weights the contribution of stereo versus monocular information, switching between detection modes to optimize performance for different ranging scenarios.

Inventive Principle:
Principle #15Dynamics

3Reliability

If detection threshold is lowered to detect uncertain objects, then detection sensitivity is improved, but false positive rate increases

Engineering Contradiction:
Improvepedestrian detection reliabilityVSAvoidfalse alarm rate
Core Design Contradiction:
ReliabilityVSObject-generated harmful factors

Solution Approach 1:

The system implements feedback through confidence level evaluation. The controller assesses the confidence level of each detected obstacle and uses this feedback to determine the appropriate warning strategy. High-confidence detections trigger immediate warnings, while low-confidence detections initiate preparatory measures without causing false alarms, thus maintaining reliability while reducing false positives.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent applies preliminary action by preparing vehicle control systems in advance for potential obstacles with uncertain detection status. When an obstacle is detected with low confidence, the system pre-adjusts control parameters and prepares emergency responses without issuing false warnings to the driver, thereby maintaining detection sensitivity while avoiding false alarms.

Inventive Principle:
Principle #10Preliminary action

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

Enables rapid emergency braking and precise vehicle control by adjusting warning and control methods according to detection reliability, improving collision prevention even when detection is uncertain, such as when pedestrians jump out of shadows.

Implementation Method 1

a method of using a camera, particularly a stereo camera

Methodology Applied
Scientific EffectParallax: Parallax

Data Source

PatentEP3188156B1Object recognition device and vehicle control system
Publication Date: 2020.05.20 HITACHI AUTOMOTIVE SYST LTD
  • EP3188156B1 patent drawingFigure 1
  • EP3188156B1 patent drawingFigure 2~3
  • EP3188156B1 patent drawingFigure 4

AI summary

The present invention addresses the problem of attaining an object recognition device that can change control of a vehicle in accordance with the reliability of detection of a target object. The object recognition device according to the present invention recognizes a target object around a vehicle and includes: a distance-information-based target object determination unit 106 that determines whether or not an object 303 is a target object by using distance information from the vehicle 301 to the object 303; an image-information-based target object determination unit 107 that determines whether or not the object 303 is a target object by using image information obtained by capturing an image of the object 303 from the vehicle 301; and a target object detection reliability calculation unit 108 that calculates the reliability of detection of a target object by using the distance information and the image information.