Performance of car image classification using the combined HOG1 and HOG2 feature extraction algorithm.

The classification of vehicle images, such as cars, using industrial automation technology can aid the police, particularly in traffic analysis. It is possible to determine cars' brand, model, and year of manufacture through their shape images. This study aims to design and analyze a system that can perform feature extraction and classification of car classes based on shape using a combined method of Histogram of Oriented Gradients 1 (HOG1) and Histogram of Oriented Gradients 2 (HOG2), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN) on the Matlab R2020a application. This study deployed car front view images, with 40 training images after pre-processing (12 Dodge Charger SRT-8 2009, 8 Dodge Dakota Crew Cab 2010, 10 Hyundai Elantra Touring Hatchback 2012, and 10 Spyker C8 Coupe 2009) and 10 testing images after pre-processing (4 Dodge Charger SRT-8 2009, 2 Dodge Dakota Crew Cab 2010, 2 Hyundai Elantra Touring Hatchback 2012, 2 Spyker C8 Coupe 2009). The research findings with the combined feature extraction of HOG1 and HOG2 unveiled that the Quadratic SVM classification yielded the highest SVM accuracy, with a maximum accuracy of 97.5% for training data and 100% for testing data. Meanwhile, the fine KNN classification, with a maximum accuracy of 95% for training data and 60% for testing data, generated the greatest KNN accuracy. [ABSTRACT FROM AUTHOR]