SlumpGuard: An AI-Powered Real-Time System for Automated Concrete Slump Prediction via Video Analysis

TL;DR

SlumpGuard introduces a fixed-camera, vision-based system for real-time concrete slump prediction using a three-stage pipeline: chute detection, pouring onset/which chute identification, and video-based slump classification. It leverages oriented bounding-box detection (YOLOv8) and optical-flow–driven timing to isolate pouring events, followed by a ResNet-3D-based video classifier with advanced augmentations and label smoothing to predict slump categories. Across a site-replicated dataset of over 6,000 clips, the approach achieves near-perfect chute localization, robust pouring detection, and slump prediction accuracy around 0.82, with expert human evaluators showing substantial subjectivity in visual slump estimation. The work demonstrates practical deployability, non-intrusive operation, and a path toward automated quality control in concrete construction workflows.

Abstract

Concrete workability is essential for construction quality, with the slump test being the most widely used on-site method for its assessment. However, traditional slump testing is manual, time-consuming, and highly operator-dependent, making it unsuitable for continuous or real-time monitoring during placement. To address these limitations, we present SlumpGuard, an AI-powered vision system that analyzes the natural discharge flow from a mixer-truck chute using a single fixed camera. The system performs automatic chute detection, pouring-event identification, and video-based slump classification, enabling quality monitoring without sensors, hardware installation, or manual intervention. We introduce the system design, construct a site-replicated dataset of over 6,000 video clips, and report extensive evaluations demonstrating reliable chute localization, accurate pouring detection, and robust slump prediction under diverse field conditions. An expert study further reveals significant disagreement in human visual estimates, highlighting the need for automated assessment.

Figures (7)

• Figure 1: An example of our collected dataset. The concrete pouring from two concrete mixer truck's chutes was captured using a ZED 2i stereo camera. The overall resolution of each camera system is $3840 \times 1080$, consisting of two individual cameras with $1920 \times 1080$ resolution each.
• Figure 2: An example of data annotation for detecting concrete pouring regions using bounding boxes. The red boxes represent the chutes, while the blue boxes indicate the corresponding unrotated bounding boxes.
• Figure 3: Overview of our pipeline. Our pipeline consists of three steps: detecting chutes, identifying when and from which chute an object falls, and predicting the slump. Finally, we compare the predicted slump with the requested range to perform anomaly detection. In the figure, "URChute" refers to an unrotated chute. The variables $x, y, w, h$, and $\theta$ represent the bounding box's center $x$-coordinates, $y$-coordinates, width, and height, respectively, with $\theta$ measured in degrees.
• Figure 4: Statistics of our dataset. The outer pie shows the distribution of slump ranges, while the inner pie details the specific slump values within each category.
• Figure 5: Qualitative results of chute detection. Blue and Sky blue bounding boxes represent detections of "URChute" and "Chute", respectively. The number inside each box denotes the confidence score.