A quality enhancement method for vehicle trajectory data using onboard images

Crowdsourcing is essential for real-time map updates. However, low-cost GPS devices often cause issues such as trajectory point deviation and key attribute missing, which reduces the potential application of trajectory data and subsequently impacts the accuracy of map updates and the quality of transportation services. Map-matching algorithms (MMs) are commonly employed to vertically match trajectory points with the road on which the vehicle is traveling, based on characteristics between the trajectory points and urban road network data. Nonetheless, challenges such as relative deviation between trajectory points, overall deviation parallel to the road direction, and abnormal vehicle heading angles still persist, increasing the difficulty of trajectory position correction. Considering the potential of onboard videos in crowdsourced data to reflect the driving environment and attitude changes of vehicles, this study proposes a trajectory quality enhancement method using onboard videos and computer vision technology to address the aforementioned issues. The experimental results indicate that the maximum difference between the estimated angle and the true angle does not exceed 35°. The estimated position can correct over 95% of the local offsets in trajectory points, and the accuracy of trajectory overall positional deviation correction is within 20 m. These findings demonstrate that the proposed method effectively addresses the impacts of angle abnormalities, trajectory point local deviations, and trajectory overall positional deviations on MMs. Furthermore, integrating the proposed method with MMs can significantly enhance trajectory position correction accuracy, which provides valuable technical support and method references for promoting the practical application of map crowdsourcing updates.