Vehicle Detection Approach Adjusting Road Curves to Estimate Local Traffic Density under Real Driving Conditions

Abstract

This paper has measured local traffic density under real driving conditions. The methodology employed is based on the traditional process of image recording. However, the conventional method has suffered from three typical limitations: long processing times, low vehicle detection performance, and inaccurate road length estimates. This study proposed two sequential approaches to resolve the three drawbacks. First, the you-only-look-once (YOLO) algorithm was used to shorten the computation time and enhance vehicle detection accuracy. Second, the vanishing point was identified during digital image processing to adjust the vertical and horizontal curves of roads, thereby calculating accurate road lengths. The proposed framework was applied to an arterial road in Seoul, South Korea. This approach was fast enough to measure local traffic density in real time. The performance of vehicle detection evaluated using a weighted mean average precision (mAP) was high at 88.73%. The accuracy of estimated road lengths evaluated using the Pearson correlation coefficient was also high at 0.97. Some co-benefits expected from the proposed method are also discussed.