Pedestrian Detection via Kernel Discriminant Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle periphery observation systems face challenges in accurately detecting pedestrians due to environmental fluctuations such as day-night changes and seasonal variations, leading to reduced efficiency and potential failure in pedestrian detection.
Innovation Solution
The implementation of an image processing equipment that performs kernel discriminant analysis on photographed images, using an input feature vector generation unit, discriminant analysis operations unit, score accumulation unit, and score judgment unit to determine if an object is a pedestrian, with features such as size, upper portion shape, and side portion shape being normalized and analyzed in a discriminant space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If infrared radiation camera is used for pedestrian detection, then detection capability in certain conditions is improved, but detection accuracy deteriorates under environmental fluctuations such as day-night changes and seasonal variations
Solution Approach 1:
The patent changes the detection parameter from thermal radiation (infrared) to optical reflection (visible light). By using visible light cameras instead of infrared cameras, the system detects pedestrians based on reflected light rather than thermal emission, making detection results independent of temperature differences between the human body and ambient atmosphere. This parameter change resolves the contradiction by maintaining detection accuracy across varying environmental conditions.
2Measurement precision
If complex image processing is performed to improve pedestrian identification accuracy, then detection accuracy is improved, but processing time increases
Solution Approach 1:
The patent segments the image processing task into distinct stages: first extracting candidate regions using simple features (size, position), then performing detailed analysis only on these candidates using kernel discriminant analysis. This segmentation allows the system to maintain high accuracy through sophisticated analysis while reducing overall processing time by limiting complex operations to a small subset of potential pedestrian regions rather than analyzing the entire image.
Solution Approach 2:
The patent performs preliminary filtering by extracting and analyzing basic features (object size, position, simple shape characteristics) before applying the more computationally intensive kernel discriminant analysis. This preliminary action identifies candidate pedestrian regions that are then subjected to detailed analysis, ensuring high accuracy while minimizing processing time by avoiding exhaustive analysis of all image regions.
3Measurement precision
If comprehensive feature analysis is performed on all detected objects, then identification accuracy is improved, but operational load increases
Solution Approach 1:
The patent applies comprehensive feature analysis (kernel discriminant analysis with multiple features including size, upper portion shape, and side portion shape) only to candidate objects that have already been identified through preliminary filtering. By performing partial analysis on the entire image set followed by excessive (comprehensive) analysis only on selected candidates, the system achieves high identification accuracy while keeping operational load manageable through selective application of complex processing.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A design for the high speed identification of a pedestrian at an image is provided, having an input feature vector generation unit 23b for generating, based on an image of an object, an input feature vector u that includes as elements the object's size, the object's upper portion shape, and the object's side portion shape; a kernel discriminant analysis operations unit 23c for generating a vector y mapped to the discriminant space, through performing operations for kernel discriminant analysis based on the input feature vector; and an object determination unit 23d for determining whether the object is a pedestrian or not, depending on whether or not this vector y is within a fixed area at the discriminant space.