Dangerous driving behavior detection using video-extracted vehicle trajectory histograms. Issue 5 (3rd September 2017)

• Record Type: Journal Article
• Authors: Chen, Zhijun; Wu, Chaozhong; Huang, Zhen; Lyu, Nengchao; Hu, Zhaozheng; Zhong, Ming; Cheng, Yang; Ran, Bin
• Is Part Of: Journal of intelligent transportation systems. Volume 21:Issue 5(2017)
• Page Start: 409
• Page End: 421
• Publication Date: 2017-09-03
• Subjects: behavior detection -- dangerous driving behaviors -- traffic safety -- trajectory histograms -- video surveillance system; Intelligent transportation systems -- Periodicals; Transportation -- Technological innovations -- Periodicals; 388.312
• DOI: 10.1080/15472450.2017.1305271
• Languages: English
• ISSNs: 1547-2450

Abstract

ABSTRACT: Dangerous driving behavior detection can be used in video surveillance systems to identify dangerous patterns, such as Abrupt Double Lane Change (ALC), Retrograde Driving (RD), and Illegal U-Turn (IT), for traffic design, traffic management, and law enforcement. The purpose of this study is to develop a detection method of dangerous driving behavior based on video surveillance. First, a novel method named trajectory histogram is proposed. A set of trajectory histograms (e.g., control points histogram and velocity histogram) is constructed to represent vehicle motion. Then, a frequently used feature selection method named Minimum Redundancy and Maximum Relevance (mRMR) is introduced to evaluate the most representative trajectory histograms for dangerous driving behavior detection. In addition, a hybrid algorithm, Particle Swarm Optimization-Support Vector Machine (PSO_SVM), is then developed to identify dangerous driving behavior. To validate the performance and effectiveness of the proposed method, several experiments are conducted. The results show that mRMR is better than other representative methods, namely Conditional Mutual Information Maximization (CMIM), Mutual Information Maximization (MIM), and ReliefF, for evaluating the most representative trajectory histograms. Based on the most representative trajectory histograms, the detection accuracy rate of dangerous driving behavior using PSO_SVM is superior to those of the most frequently used classifiers—Naïve Bayesian Classifier (NBC), k-Nearest Neighbor (kNN), and C4.5 decision tree. In addition, we find that the proposed method outperforms the two common approaches for dangerous driving behavior detection in video surveillance systems. Therefore, the proposed method can be widely applied to detect dangerous driving behavior in video surveillance systems.