Assessing Pavement Rougness Using Jitter Vector from In-vehicle Camera Videos

The process of assessing pavement smoothness is cumbersome, inefficient and time-consuming. To address these issues, a pavement smoothness assessment method based on in-vehicle video jitter vectors is proposed. This method enables preliminary and rapid screening of pavement conditions under normal scenarios. It uses driving videos collected by onboard devices as the assessment data. Preprocessing enhances the contrast of driving video images and reduces the effect of changes in the driving environment on the contrast of video images. The video images then undergo block-wise grayscale projection and similarity determination to remove significant deviations in jitter vectors and interference from moving objects. This extracts the main jitter vectors from the driving videos. The particle swarm optimization (PSO) algorithm improves the search pattern of the projection correlation curve. Using the grayscale projection curve correlation formula as the fitness function in the row (or column) direction enhances search efficiency of the algorithm. A genetic algorithm (GA) optimized K-means clustering algorithm is established to autonomously assess road smoothness at different vehicle speeds by combining vehicle speed and video jitter vectors. Experimental validation shows that the PSO-based grayscale projection algorithm detects smooth road surfaces in 0.148 s, improving efficiency by 91.41% compared to the original algorithm. For rough road surfaces, detection takes 0.123 s, improving efficiency by 87.58%, and consistently detects jitter vector values. The GA-K-means algorithm effectively reduces interference from initial cluster centers, avoiding premature conver-gence.