Machine Learning Approach for Identification of Accident Severity from Accident Images Using Hybrid Features

Abstract

Rapid growth in automobiles has caused an upsurge of accidents per day, which leads to the loss of lives and incurable disabilities to the victims. Therefore, the severity of the accident must be analyzed in real-time to save the injured and enhance emergency services. Accordingly, the accident image is considered as significant data in this work. From the accident image, essential features such as shape, texture and intensity gradient features are extracted using Hu moments, Local Binary Pattern (LBP) and Histogram of Oriented Gradients (HoG) respectively. The extracted image features are combined to form a hybrid feature vector. With an objective to recognize the severity of the accident, the hybrid feature is employed to train the machine learning classifier models such as Support Vector Machine (SVM), Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), AdaBoost (AB) and Gradient Boosting (GB). The performance of the classifiers is evaluated in terms of Area under the curve (AUC), precision, recall and F1-score. The results show the Random Forest performs better with AUC 0.75 compared to other models. Moreover, the result also reveals that hybrid features improve the recognition rate compared to the single feature.