A Vehicle Detection Method Based on Convolutional Neural Network Optimized by Genetic Algorithm

This manuscript introduces an advanced vehicle target detection algorithm which leverages the integration of improved underside shadow characteristics with a Convolutional Neural Network (CNN) optimized by genetic algorithms. Our approach augments underside shadow delineation through difference enha...

Full description

Saved in:
Bibliographic Details
Published in:International Conference on Intelligent Control and Information Processing (Online) pp. 227 - 233
Main Author: Sun, Qingqing
Format: Conference Proceeding
Language:English
Published: IEEE 08.03.2024
Subjects:
Convolutional neural networks
Feature extraction
Genetic Algorithm
Hypothetical Vehicle
Interference
Lighting
Object detection
Roads
Support Vector Machine
Vehicle detection
ISSN:2835-9577
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This manuscript introduces an advanced vehicle target detection algorithm which leverages the integration of improved underside shadow characteristics with a Convolutional Neural Network (CNN) optimized by genetic algorithms. Our approach augments underside shadow delineation through difference enhancement techniques to amplify the grayscale contrast between shadows and the road surface. This, in turn, facilitates the precise extraction of road boundaries, streamlining the computational process when generating potential vehicle shadow candidates. By harnessing both the geometric and textural properties of the vehicle shadows, our algorithm eficiently sifts through and formulates viable vehicle presumptions. The innovative application of a genetic algorithm to fine-tune the CNN addresses the common pitfalls of local optima encountered with conventional CNN training methodologies, such as the gradient descent approach, which often initializes weights randomly. This optimization yields a substantial increment in the accuracy of vehicle detection. The algorithm's superior performance is further substantiated by comprehensive validation on a vast collection of real-world roadway images within MATLAB. This rigorous testing confirms the algorithm's adeptness in differentiating between genuine vehicle underside shadows and other interfering regions, consequently ensuring exceptionally reliable vehicle detection. The system shows remarkable adaptability and robustness across a variety of complex environments, characterized by fluctuating illumination, multiple tra fic lanes, and vehicles exhibiting diverse positions and orientations.
ISSN:2835-9577
DOI:10.1109/ICICIP60808.2024.10477793