A Real-Time Warning System for Rear-End Collision Based on Random Forest Classifier

Abstract

Rear-end collision warning system has a great role to enhance driving safety. In this system, some measures are used to evaluate the safety and in the case of dangerous, the system warns drivers. This system should be executed in real-time, to remain enough time to avoid collision with the front vehicle. To this end, in this paper, a new system is developed by using a random forest classifier to extract knowledge about warning and safe situations. This knowledge can be extracted from accidents and vehicle trajectory data. Since the data of these situations are imbalanced, a combination of cost-sensitive learning and classification methods was used to improve the sensitivity, specificity, and processing time of classification. To evaluate the performance of this system, vehicle-trajectory-data of 100 cars that have been provided by Virginia tech transportation institute, are used. The comparison results are given in terms of accuracy and processing time. By using TOPSIS multi-criteria selection method, it is shown that the implemented classifier is better than different classifiers including Bayesian network, Naive Bayes, MLP neural network, support vector machine, k-nearest neighbor, rule-based methods and decision tree. The implemented random forest gets 88.4% accuracy for detection of the dangerous situations and 94.7% for detection of the safe situations. Also, the proposed system is more robust compared with the perceptual-based and kinematic-based algorithms.