ConcreteVision | Algotechniq

Motivation

The Concrete Layer Coverage Interval (CLCI) is the critical time between when concrete is poured and when it needs to be covered to maintain proper moisture and temperature for curing.

Proper coverage prevents issues like cracking and weakening due to exposure to environmental factors. Overtime alerts for CLCI are essential because they notify construction teams if the concrete isn’t covered in time, helping to preserve its structural integrity and durability. This real-time alert system ensures that the concrete cures correctly, reducing the risk of defects and ensuring the safety and quality of the structure.

Objective

The project introduces a new approach for monitoring the progress of concrete pouring in construction using deep object detection and semantic segmentation models. The specific novelty lies in achieving high accuracy for overtime alerting of concrete layer coverage intervals, which improves the precision and timeliness of construction monitoring.

Algotechniq's AI Vision Transformer Monitoring Concrete Flow.

Prior Work

The monitoring of concrete pouring progress and quality control was primarily done manually. This manual approach often led to issues such as insufficient timeliness, misreporting, and omission of critical data, which could negatively impact the overall construction quality.

Benefits

Algotechniq’s technology will benefit construction management teams by providing more accurate, real-time data on construction progress. This improvement supports better quality control, progress tracking, and overall productivity analysis in construction projects, particularly in large-scale infrastructure projects.

Enabling Technologies

The key enabling technologies include deep learning, computer vision techniques such as semantic segmentation and object detection, and advanced data processing methods like data augmentation and model training on specialized datasets. The use of Vision Transformers (ViTs) (Arnab et al., 2021), pre-trained on large datasets and fine-tuned for specific construction scenarios, is central to this approach.

Quantifying Performance

The success of the new approach can be quantified by the accuracy and timeliness of the overtime warnings for concrete layer coverage intervals. Specifically, the approach is evaluated by comparing the model’s predictions with manually annotated real-world data, ensuring that the model can accurately detect the construction progress and provide timely warnings to prevent delays.

We present pure-transformer based models for video classification, drawing upon the recent success of such models in image classification. Our model extracts spatiotemporal tokens from the input video, which are then encoded by a series of transformer layers. In order to handle the long sequences of tokens encountered in video, we propose several, efficient variants of our model which factorise the spatial- and temporal-dimensions of the input. Although transformer-based models are known to only be effective when large training datasets are available, we show how we can effectively regularise the model during training and leverage pretrained image models to be able to train on comparatively small datasets. We conduct thorough ablation studies, and achieve state-of-the-art results on multiple video classification benchmarks including Kinetics 400 and 600, Epic Kitchens, Something-Something v2 and Moments in Time, outperforming prior methods based on deep 3D convolutional networks.