Bump feature detection of the road surface based on the Bi-LSTM

Abstract

Abstract The road network is the basic facility for transportation systems in the city. Every day, a large number of vehicles move on the road and exert different pressure on the ground, which leads to various problems for the road surface, such as the bump features of the road surface (BFRS). However, traditional methods, such as detecting BFRS manually or with professional equipment, require a lot of professional management and devices. Based on the mobile sensor and the bidirectional long short-term memory (Bi-LSTM), a detection method for BFRS is proposed. The BFRS detection method proposed in this article solves the problem that other BFRS detection methods cannot detect large area road surface efficiently and provides an algorithm idea for efficient detection of large area road surface BFRS. The mobile phone with multi-sensors is carried on vehicles, and the BFRS information is logged during the movements. The orientation of the mobile is computed according to the gyroscope. The actual posture of the acceleration sensor is adjusted with the reference coordinate system, whose z-axis is vertical to the ground. This article uses the adjusted acceleration data as the training dataset and labels it according to time stamps and videos recorded by the driving recorder. Finally, the Bi-LSTM is constructed and trained, followed by the BFRS detection. The results show that it can detect BFRS in different regions. The detection accuracy of the campus section and the extended experiment was 92.85 and 87.99%, respectively.