In-situ concrete slump test incorporating deep learning and stereo vision

Measuring the concrete slump is necessary for ensuring the quality of concrete before its use. However, manual slump measurements have limited accuracy and are time-consuming and labor-intensive. Herein, an approach based on deep learning and stereo vision techniques is proposed to effectively improve the determination of concrete slump in outdoor environments. Input images obtained from a stereo camera system were classified into three slump cases via deep learning. The actual slumps were then calculated using depth maps obtained via stereo vision. The results were analyzed for the correlations between camera mounting heights and the distance between the cameras to identify the optimal settings for an onsite working system. A baseline of 70 mm and working height of 1.7–1.9 m yielded optimal results with errors ≤2.05%. Additionally, the mask-region-based volumetric results exhibited errors of <8.9% relative to ground truth, thereby validating the reliability of the proposed approach.