<p>The road longitudinal profile plays a vital role in evaluating road roughness, as it directly influences vehicle comfort and safety. Although accelerometer-based measurement methods are cost-effective and convenient, their accuracy and stability are often undermined by acceleration fluctuations during variable-speed motion. Additionally, these methods face difficulties in managing complex road surface undulations, limiting their precision and efficiency in contemporary infrastructure maintenance. To address these limitations, this study proposes a high-precision road longitudinal profile measurement algorithm that integrates dual-point laser data compensation, Euler angles, and a Transformer deep learning model to minimize errors. The algorithm incorporates Butterworth filtering to eliminate laser noise and combines Inertial Measurement Unit (IMU) and dual-point laser data to calculate Euler angle compensation, leveraging the relative laser height differences and installation distances to enhance the representation of road surfaces. The Transformer model, equipped with a self-attention mechanism and multi-layer perceptron, refines the measurement process by utilizing transfer learning on acceleration data and measured profiles to mitigate acceleration-induced interference. Experimental results indicate that the proposed method outperforms traditional techniques in terms of accuracy and robustness under variable speeds and conditions. This research advances the use of deep learning in transportation engineering, offering an efficient and precise solution for infrastructure maintenance. It significantly contributes to the automation of road inspections and addresses critical challenges in modern transportation systems.</p>