Classification of Road Accidents Using SVM and KNN

Abstract

The rapid increase in automobiles in the recent past has caused an unrestrained escalation of road accidents. Due to road accidents, victims suffer from non-fatal injuries and incurring disabilities. The road accidents have tended many researchers to analyze the severity and type of the accident to enhance road safety measures and aid to speed up the post-crash support amenities. This paper attempts to classify the severity of the accidents by analyzing the accident images. The features of the accident images are extracted using algorithms such as histogram of oriented gradient (HOG), local binary pattern (LBP) and speeded up robust features (SURF). These features are given as input to k-nearest neighbor (KNN) and support vector machine (SVM) to classify the severity of the accidents. The performance of SVM and KNN classifiers with three feature extraction algorithms is assessed and compared. The classification results show that SVM classifier outperformed KNN. SVM with the HOG features shows better accuracy of 79.58% compared to LBP and SURF.