Pedestrian Recognition Apparatus for Vehicle Collision Avoidance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current around view monitoring systems do not effectively recognize pedestrians or determine collision possibilities with vehicles, lacking the ability to provide visual guidance and autonomous emergency braking.
Innovation Solution
A pedestrian recognition apparatus that uses image processing to detect and classify pedestrians based on mobility and distance from a vehicle, performing HOG and SVM operations to determine collision risks and overlay visual guidance, while controlling vehicle velocity or generating warning sounds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pedestrian recognition is added to around view monitoring system, then safety is improved, but device complexity increases
Solution Approach 1:
The pedestrian recognition process is divided into distinct modules: object extraction unit, pedestrian candidate detection unit, and pedestrian possibility determination unit. Each module performs a specific function (extracting objects from images, detecting pedestrian candidates based on vertical components, determining pedestrian possibility using HOG and SVM), allowing the complex recognition task to be managed through modular, independent components that can be developed and optimized separately.
2Reliability
If real-time pedestrian detection is implemented, then collision prevention is improved, but processing time increases
Solution Approach 1:
The object extraction unit isolates potential pedestrian regions from the entire camera image by identifying and extracting objects based on vertical components. This extraction step reduces the amount of data that needs to be processed by subsequent pedestrian detection algorithms, focusing computational resources only on relevant regions rather than analyzing the complete image, thereby reducing overall processing time.
Solution Approach 2:
The system applies simplified detection methods (vertical component analysis) for initial pedestrian candidate identification, and then applies more computationally intensive methods (HOG and SVM) only to the extracted candidate regions rather than the entire image. This partial application of processing steps reduces overall computational load while maintaining detection accuracy.
3Measurement precision
If multiple image processing operations (HOG, SVM) are performed, then pedestrian detection accuracy is improved, but computational load increases
Solution Approach 1:
The object extraction unit performs preliminary processing by identifying and extracting potential pedestrian regions before applying the computationally intensive HOG and SVM operations. This preliminary action filters out non-pedestrian regions, ensuring that expensive computational operations are applied only to relevant candidate regions, thereby reducing overall energy consumption while maintaining high detection accuracy.
Data Source
AI summary
The present invention relates to a pedestrian recognition apparatus. A pedestrian recognition apparatus may comprises: an image receiving unit configured to sequentially receive a plurality of images from a camera; and a pedestrian determination unit configured to perform a pedestrian candidate object detection process of detecting a pedestrian candidate object from one or more objects in the images, a mobility determination process of determining whether the pedestrian candidate object is moving, using the plurality of images, and setting the pedestrian candidate object to a moving pedestrian candidate object, and a pedestrian possibility determination process of performing a predefined operation on the moving pedestrian candidate object, and setting the moving pedestrian candidate object to a pedestrian when a value calculated through the predefined operation is equal to or more than a threshold value.