Airfield concrete pavement joint detection network based on dual-modal feature fusion

To solve the problem of inaccurate positioning of airfield pavement concrete joint and misalignment quantification, this study proposes an airfield concrete pavement joint detection network based on dual-modal feature fusion (ACJD-DFF) and a misalignment quantification method. First, outlier cleaning and row-level distortion correction are proposed to improve the quality of 3D data. Subsequently, data from two modalities are fused to create a dual-modal feature fusion matrix dataset. An ACJD-DFF is proposed to accomplish the positioning of the concrete joint by integrating positioning coordinate optimization. Finally, by mapping the accurate joint positioning coordinate by ACJD-DFF, the misalignment quantification of the concrete joint is constructed. The results show that the ACJD-DFF performs better than the state-of-the-art methods. The mAP@0.5 and Recall reached 96.38%, 91.4%. The average error of misalignment quantification is 4.333%. These methods have been applied in routine airfield pavement monitoring and are important for later detection of concrete joint distress. •A high precision surface data collecting robot (HPSCR) has been developed for concrete joint detection tasks.•The outlier point cleaning and row-level distortion correction were proposed to restore the pavement details more accurately.•The ACJD-DFF integrates positioning coordinate optimization to make the detection completely contain the entire joint area.•A misalignment quantification was developed based on the positioning coordinates to measure the 9 evenly distributed points.