Vehicle Detection Using Edge-Based Candidate Group Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle detection systems using camera images struggle to accurately detect preceding vehicles under varying illumination and weather conditions, particularly when the side or bottom surfaces are unclear, leading to potential safety threats due to inadequate warnings or braking.
Innovation Solution
A vehicle detection system that generates a vehicle candidate group by presuming the side or bottom surface candidate groups using edge detection and binarization techniques, even when one is not detected, and filters these groups to accurately identify preceding vehicles, incorporating a camera, edge detection unit, candidate group detection units, and a vehicle detection unit with weak and strong classifiers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a camera-only system is used for vehicle detection, then system complexity and production cost are reduced, but detection reliability deteriorates under varying illumination and weather conditions
Solution Approach 1:
The detection process is segmented into multiple independent stages: edge detection, binary image generation (horizontal and vertical), candidate group detection, and final vehicle detection. This segmentation allows each stage to focus on specific features, improving overall reliability while keeping the system structure relatively simple.
Solution Approach 2:
The system transforms the detection problem from a single 2D image analysis to a multi-dimensional approach by generating both horizontal and vertical binary images from edge detection. This dimensional transformation enables detection of vehicle features that may be obscured in the original image, improving detection reliability under varying conditions.
2Measurement precision
If strict detection of all vehicle surfaces is required, then measurement precision is improved, but detection capability deteriorates when side or bottom surfaces are not visible
Solution Approach 1:
The system generates candidate groups for side surfaces and bottom surfaces separately, then combines them to form vehicle candidate groups. This partial action approach allows the system to detect vehicles even when only some surfaces are visible, as the detection can proceed with available information rather than requiring complete surface detection.
Solution Approach 2:
The system performs preliminary detection of side surface candidate groups and bottom surface candidate groups before combining them into vehicle candidate groups. This preliminary action allows the system to identify potential vehicle features independently, then synthesize them into complete vehicle detections even when some features are partially occluded.
3Device complexity
If conventional detection methods are used, then system simplicity is maintained, but detection precision deteriorates when vehicles are not perfectly photographed
Solution Approach 1:
The system introduces intermediate processing stages including edge detection and binary image generation as mediators between the raw camera image and final vehicle detection. These intermediary steps enhance detection precision by extracting and emphasizing relevant features while filtering out noise, without significantly increasing overall system complexity.
Data Source
AI summary
The invention relates to a vehicle detection method which includes: photographing a front-view of a vehicle; detecting an edge from an image photographed by a camera and generating a horizontal binary image and a vertical binary image; detecting a bottom surface candidate group relating to a preceding vehicle from the horizontal binary image and detecting a side surface candidate group relating to the preceding vehicle from the vertical binary image; when one of a pair of side surface candidate groups and bottom surface candidate groups of the pair of side surface candidate groups is not detected, estimating the side surface candidate group and bottom surface candidate group and generating a vehicle candidate group; and filtering the vehicle candidate group and detecting the preceding vehicle. Therefore, even when all of vehicles are not perfectly photographed, a preceding vehicle can be more precisely detected.