Vehicle Object Detection Using Segmented Optical Flow Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems for vehicles struggle to accurately detect objects, such as forward vehicles, without relying on large volumes of teacher data and often fail to identify unique vehicle shapes or objects moving towards the vanishing point in camera images.
Innovation Solution
An object detection apparatus that divides the search region into upper and lower regions using optical flow models with a fixed magnification factor and planar constraint conditions, determining a boundary line that minimizes pixel value differences to differentiate between vehicle and road surface regions, allowing for accurate object detection without pre-stored teacher data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a single optical system camera is used for object detection, then the system cost is reduced, but the detection accuracy deteriorates
Solution Approach 1:
The patent divides the image into multiple search regions (first search region for vehicles, second search region for pedestrians) with different detection parameters. This segmentation allows the single-camera system to achieve detection accuracy comparable to multi-camera systems by applying region-specific optimization strategies.
Solution Approach 2:
Different detection parameters and criteria are applied to different regions of the image. The first search region uses parameters optimized for vehicle detection while the second search region uses parameters optimized for pedestrian detection, thereby improving overall detection accuracy without adding hardware complexity.
2Productivity
If optical flow model with fixed magnification factor is applied to the entire search region, then the processing speed is improved, but the detection accuracy for objects at different distances deteriorates
Solution Approach 1:
The patent segments the search region into multiple zones (first search region and second search region) and applies different optical flow model parameters to each zone. This allows the system to maintain processing speed while improving detection accuracy for objects at varying distances by using distance-appropriate magnification factors.
Solution Approach 2:
Different magnification factors are assigned to different spatial regions based on object distance. The first search region uses a first magnification factor appropriate for closer objects while the second search region uses a second magnification factor for farther objects, thereby maintaining both processing efficiency and detection precision across varying distances.
3Productivity
If teacher data is stored in advance for image recognition, then the detection speed is improved, but the ability to detect unique vehicle shapes deteriorates
Solution Approach 1:
The system performs self-learning by automatically acquiring detection parameters from images captured during normal operation. Instead of relying on pre-stored teacher data, the system adapts to unique vehicle shapes and environmental conditions by learning from actual operational data, thereby maintaining both detection speed and adaptability to novel objects.
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
This approach improves the accuracy of object detection by optimizing the optical flow models for both vehicle and road surface regions, enabling the detection of objects like forward vehicles without pre-stored data, reducing erroneous detections, and providing timely alerts for collision prevention.
Implementation Method 1
an image captured by a camera
Data Source
AI summary
A vehicle detection apparatus detects an object based on an image captured from a vehicle. In the vehicle detection apparatus, a camera is mounted in the vehicle. A search region determining unit determines a region including a road surface in the image to be a search region. A cost minimizing unit divides the search region into two, an upper region and a lower region, by a boundary line, determines corresponding points in a plurality of images captured by the camera at different time points by applying an optical flow model having a fixed magnification factor to the upper region and an optical flow model having a planar constraint condition to the lower region, and determines a boundary line that minimizes cost related to a difference between pixel values of the corresponding points. A vehicle presence determining unit determines presence of a forward vehicle based on the boundary line.