PURPOSES : This study aims to develop and evaluate computer vision-based algorithms that classify the road roughness index (IRI) of road specimens with known IRIs. The presented study develops and compares classifier-based and deep learning-based models that can effectively determine pavement roughness grades.
METHODS : A set road specimen was developed for various IRIs by generating road profiles with matching standard deviations. In addition, five distinct features from road images, including mean, peak-to-peak, standard variation, and mean absolute deviation, were extracted to develop a classifier-based model. From parametric studies, a support vector machine (SVM) was selected. To further demonstrate that the model is more applicable to real-world problems, with a non-integer road grade, a deep-learning model was developed. The algorithm was proposed by modifying the MNIST database, and the model input parameters were determined to achieve higher precision.
RESULTS : The results of the proposed algorithms indicated the potential of using computer vision-based models for classifying road surface roughness. When SVM was adopted, near 100% precision was achieved for the training data, and 98% for the test data. Although the model indicated accurate results, the model was classified based on integer IRIs, which is less practical. Alternatively, a deep-learning model, which can be applied to a non-integer road grade, indicated an accuracy of over 85%.
CONCLUSIONS : In this study, both the classifier-based, and deep-learning-based models indicated high precision for estimating road surface roughness grades. However, because the proposed algorithm has only been verified against the road model with fixed integers, optimization and verification of the proposed algorithm need to be performed for a real road condition.
Since the road management paradigm has changed into the user-oriented circumstance, the functionality of the crucial road maintenance factors became important than before. Among these factors, the roughness directly related to the ride quality for driver became to get more attention. IRI(International Roughness Index) is recently the most widely used roughness indices in the world. IRI is a reasonable index that reflects the vertical displacement(bounce) of vehicle as the road profile changes. Since IRI reflects the vertical behavior of vehicle, it reflects ride quality indirectly. However, there are various rotational behaviors such as roll, yaw, and pitch in addition to the vertical displacement. Profiles, which MRI range was 1.13-4.12m/km, were measured in five sections and the profiles were entered into CarSIM to simulate vehicle behavior. As a result, the pitch was the largest in all sections, followed by roll and yaw, relatively. Especially, the amount of yaw is about 5% of the pitch or about 7% of the roll. The behavior of moving vehicle was measured using INS(inertial navigation system) and accelerometer in the section where the road surface profile was measured. As a result, as in the simulation, the pitch was the largest in all the sections and the amount of yaw is only about 7% of the pitch or about 18% of the roll. Field experiments were conducted to analyze the effect of the rotational behavior of the actual driving vehicle on the ride quality. 33 panels evaluated the ride quality on a ten-point scale while driving on 35 sections with various roughnesses. 35 test sections were selected considering the roughness distribution of actual expressway. The panel was selected considering age, driving experience, gender, and expertise. Of the total 1,155 responses, 964 responses were used for the analysis, except 191 responses measured at low driving speeds. In addition, the amount of vehicle behavior and road surface profile were measured using INS and laser. As a result of correlation analysis between MPR(mean panel rating) and vehicle behavior, correlation coefficient of bounce was the highest with 0.814, and the order of pitch was 0.798, and roll was 0.734, relatively. As a result of regression analysis for predicting ride quality, regression model combining bounce and roll was statistically the most suitable. This model is expected to reflect the ride quality more effectively because it can consider the vehicle behavior due to the longitudinal profile change of the road surface as well as the vehicle behavior due to the difference between the left and right wheel path road profile.