Abnormal driving trace detection method of intelligent vehicles based on alignment analysis

Road alignment is one of the main factors affecting traffic safety. The vehicle driving trace is the main basis of road alignment design, which has the geometric characteristics of ride comfort, continuous trace curvature, and continuous curvature variation. Since the geometric characteristics of alignment are less considered in the current abnormal driving trace detection methods, a multi-measure fusion method for abnormal driving trace detection of intelligent vehicles was proposed. The three evaluation indexes of curve similarity, ride comfort, and number of consecutive lane changes were integrated to comprehensively evaluate the abnormal driving trace. A piecewise-weighted-average discrete Fréchet distance (PDFD) algorithm was proposed to calculate the curve similarity. This paper analyzed the two dimensions of ride comfort, i.e., the composition of alignment components and the geometric characteristic parameters of alignment components of the driving trace. The five-point orthogonal fitting curvature method was used to extract the alignment components of the driving trace and the segmentation points of each alignment component. According to the principle of orthogonal least squares, the mathematical model of the alignment component of the driving trace was established, and the geometric characteristic parameters were calculated, such as the starting point, end point, radius of the circular curve, center of the circle, starting radius and ending radius of the transition curve, length of alignment components, and curve trend. The experimental results show that the overall accuracy of the abnormal driving trace detection method is 80.4%; the proportion of anomalies caused by the jump of the curvature variation is 41%; the accuracy of the PDFD algorithm is 81%, which is 14% higher than that of the conventional Fréchet distance algorithm of 67%.