주행 안전성 평가를 위한 블랙박스 영상 인식 적용 기법 연구

초록

To evaluate traffic safety, traffic accident data, navigation systems, driving recorders, and simulations are commonly used. However, these methods have limitations in identifying situations immediately before accidents occur. In contrast, black box camera footage can be obtained with the vehicle owner's consent and, with a 100% penetration rate, allows for comprehensive analysis of all vehicles. This study aims to analyze vehicle behavior in rear-end collision situations based on black box camera footage and to establish foundational data for future collision avoidance and response strategies for autonomous vehicles. To analyze rear-end collision black box camera footage, an image recognition algorithm based on OpenCV's YOLO5 is developed to perform object recognition in the footage. Additionally, algorithms are developed to predict the speed of preceding and following vehicles and to estimate the inter-vehicle distance. Real road driving experiments were conducted to train and calibrate the developed algorithms. Speeds and Time-to-Collision were extracted from 12 rear-end collision cases that occurred on highways. These 12 accidents were categorized into three groups based on speed, and the characteristics of each group were analyzed. Through this study, an algorithm was developed to analyze vehicle behavior immediately before rear-end collisions using black box camera footage. The further refinement of this algorithm and evaluation metrics is expected to advance video-based traffic accident analysis and contribute to the development of strategies for improving traffic safety in real road conditions.

키워드