Pedestrian Detection Fusion Using Camera-Radar Matching Areas
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional pedestrian determination systems using cameras and radars are slow due to the need to recognize shapes of multiple objects before identifying pedestrians, leading to delayed pedestrian detection.
Innovation Solution
A method that uses a camera and radar as external environment recognition sensors to detect pedestrians by setting a matching area based on the detected pedestrian candidate and expanding it when the angular difference between camera and radar detection angles is within a predetermined value, allowing for quicker pedestrian identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional radar object recognition is used to identify pedestrians, then object detection capability is improved, but pedestrian determination time increases
Solution Approach 1:
The patent applies preliminary action by pre-defining a matching area based on camera-detected pedestrian candidate position before radar object recognition is completed. This allows the system to prepare the search region in advance, so when radar data arrives, only objects within this pre-defined area need to be checked, significantly reducing pedestrian determination time while maintaining detection accuracy
2Quantity of substance
If multiple objects are detected by radar, then object detection coverage is improved, but pedestrian identification complexity increases
Solution Approach 1:
The patent applies local quality by creating a localized matching area with specific spatial boundaries around the pedestrian candidate position. This restricts the complex object recognition process to only those objects within the local matching area, rather than analyzing all detected objects system-wide, thereby reducing identification complexity while maintaining comprehensive detection coverage
Solution Approach 2:
The patent segments the object recognition process into two independent stages: camera-based pedestrian candidate detection and radar-based object detection within a defined matching area. This segmentation allows each sensor to operate independently and efficiently, with the matching area serving as a bridge that connects the two processes without requiring complex integration of all detected objects
3Measurement precision
If camera shape recognition is used to detect pedestrian candidates, then pedestrian detection accuracy is improved, but detection responsiveness decreases
Solution Approach 1:
The patent merges camera and radar detection systems into a coordinated pedestrian determination process. The camera provides accurate pedestrian candidate positions through shape recognition, while the radar simultaneously detects objects within a pre-defined matching area around that position. This merging allows the system to leverage the camera's accuracy without waiting for complete radar analysis of all objects, improving responsiveness while maintaining detection accuracy
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 pedestrian detection responsiveness by utilizing the camera's shape recognition capabilities and the radar's object recognition capabilities, enabling faster and more accurate pedestrian determination even at longer distances.
Implementation Method 1
a radar which detects an object in a forward position of the host vehicle based on reflected waves
Data Source
Figure 1
Figure 2
Figure 3
AI summary
To improve pedestrian determination responsiveness, when determining a pedestrian using a camera and a radar during travel. External environment recognition sensors that acquire forward information for a host vehicle (A) are provided, to determine that a pedestrian (B) is present in a forward position of the host vehicle (A) using the external environment recognition sensors. In this pedestrian determination method, a front camera unit (1) and a millimeter wave radar (2) are provided as the external environment recognition sensors. When a pedestrian candidate (B') is detected in a forward position of the host vehicle (A) based on an image signal from the front camera unit (1), a matching area (C) is set with the position of the detected pedestrian candidate (B') as the center point. When the position of the object (D1) that is closest to the pedestrian candidate (B') from among the plurality of objects (D1, D2, D3) detected by reflected waves from the millimeter wave radar (2) is within the matching area (C) thereby establishing a match, the pedestrian candidate (B') is determined to be a pedestrian (B).