AI inventory of construction materials from photos: turnkey development

The problem of manual counting on construction sites

Manual inventory of materials on a construction site is a source of errors and losses. Rebar, bricks, cement bags, pipes, plywood sheets — all of this can be counted automatically from a phone or drone camera photo. We develop turnkey computer vision systems for AI counting of construction materials: from dataset collection to integration with your WMS or 1C. Clients save up to 80% of time on inventory and eliminate the human factor. Average budget savings reach 1.5 million rubles per year per warehouse.

How does an AI system handle dense stacks?

Classical detectors (e.g., YOLOv8) fail when objects lie close together: IoU > 0.6, NMS suppresses correct detections. Accuracy drops to 60–70%. For such cases we use density estimation — an approach borrowed from crowd counting. The network predicts a density map whose pixel sum ≈ number of objects. For bags in stacks, accuracy rises to 91–95%. Density estimation outperforms YOLO detection by 1.3–1.5 times in accuracy on dense stacks. Density estimation is a method adapted for industrial accounting.

class DensityBasedCounter:
    """
    CSRNet or CrowdCounting approach adapted for materials.
    Instead of detecting each object, we predict a density map.
    The sum of the density map pixels ≈ number of objects.
    """
    def __init__(self, model_path: str):
        import torchvision.models as models
        # VGG16 backbone + density head
        self.model = self._build_csrnet()
        self.model.load_state_dict(torch.load(model_path))
        self.model.eval()

    @torch.no_grad()
    def count(self, image: np.ndarray) -> dict:
        from torchvision import transforms
        transform = transforms.Compose([
            transforms.ToTensor(),
            transforms.Normalize([0.485, 0.456, 0.406],
                                  [0.229, 0.224, 0.225])
        ])

        tensor = transform(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
        tensor = tensor.unsqueeze(0)

        density_map = self.model(tensor)[0, 0].numpy()
        # The sum of the density map = number of objects
        count = float(density_map.sum())

        return {
            'method': 'density_estimation',
            'count': round(count),
            'count_float': count,
            'density_map': density_map
        }

Why is density estimation more accurate than a detector for bags?

For bulk materials and objects in dense rows, a classical detector fails. YOLO detection loses to density estimation: accuracy drops to 60–70% vs. 91–95% for the density method. The density map approach yields more stable results on overlapping objects. It is based on the CSRNet architecture, originally developed for people counting in images, but also works well for construction materials.

How do we collect and label the dataset?

Training requires 1000–3000 labeled photos of materials in different angles and conditions. We help organize the shooting or accept ready images. The more diverse the dataset, the higher the accuracy. Labeling is done manually by qualified operators using polygon annotation. We create a separate training set for each material — bricks, rebar, bags, pipes. This allows the model to accurately count objects even under partial occlusion.

Case study: rebar warehouse inventory (our client)

A major construction materials distributor with a warehouse of 800 tons of rebar in bundles, 6 standard sizes (d8 to d32). Manual inventory: 2 people, 8 hours per month. After implementation: an operator photographs the ends of bundles with a tablet (20–30 photos per hour). The model counts the number of rods in each bundle: Hough circles for d12–d25, density estimation for d8.

• Count accuracy: ±2% of manual recount
• Inventory time: 1.5 hours vs. 8 hours manually
• Integration with 1C: automatic stock updates
• Annual savings: over 1.5 million rubles

How to set up AI counting for your materials?

The implementation process includes several stages. First, we audit your warehouse and material types. Then we organize photo shooting from different points and label the dataset. After training the model and testing on a control sample, we deploy the solution on your server or edge device. Integration with the accounting system is the final step. At each stage we provide an accuracy report and recommendations for fine-tuning.

What is included in the development of an AI counting system

• Process audit: analysis of current accounting methods, material types, and shooting conditions
• Dataset collection and labeling: 3000+ images with polygon annotation
• Model training: selection of architecture (YOLOv8, CSRNet, SAM) and hyperparameters
• API development: REST or gRPC for integration with your WMS/1C
• Testing: A/B test on real data, accuracy report
• Deployment: server (Docker, Kubernetes) or edge (Jetson, Raspberry Pi)
• Documentation and training: operator manual, 1-hour session
• Support: 3 months of warranty maintenance

Common difficulties during implementation

• Lighting variety: from bright sun to warehouse twilight — dataset augmentation is required.
• Occlusion and dirt: dirt on objects reduces accuracy — we use preprocessing and filtering.
• Different batches: brick shades may vary — the model must generalize across color variations.

Timeline and cost

Development cost is calculated individually and depends on the number of material types, data quality, and integration complexity. The investment typically pays back in 6–12 months due to reduced labor hours and fewer errors. We guarantee accuracy of at least 90% on the pilot project. Contact us for a consultation — our engineer will assess your task and propose the optimal solution. Order a free preliminary evaluation to discuss your project details.

How Distribution Shift Kills CV Model Metrics in Industry

On a production line, a camera is installed to control product quality. The model is trained on 10,000 labeled images—test accuracy mAP 0.84. Deployed to production, and in the first week it misses 30% of defects. Lighting on the line changes between shifts; distribution shift nullifies the metrics. This is a classic story with computer vision in industry, where pattern recognition fails without proper drift handling.

Our engineers, with experience from 60+ computer vision projects, know how to eliminate such scenarios. We guarantee stable model performance under real conditions.

Object Detection: YOLO, RT-DETR, and Everything in Between

YOLO is the standard for real-time detection. YOLOv8 and YOLOv11 from Ultralytics are the most used versions in production: simple API, active community, built-in validation, and export to ONNX/TensorRT. For tasks with high accuracy requirements and less critical latency, RT-DETR, a transformer-based architecture without NMS, gives better mAP on COCO at comparable speed to YOLOv8l.

A typical mistake when training a detector: dataset of 8000 images, 3 classes, fine-tune YOLOv8m—F1 0.73 on validation. Look at confusion matrix—one class is almost never detected. Cause: imbalance 1:23. Solution: oversampling rare class, focal loss for objectness, augmentations (Mosaic, MixUp disabled for rare class as they "blur" it). Transfer learning is mandatory: pretrained on COCO weights reduces data requirement by 10 times. Fine-tuning on 500–2000 domain images yields a working model in 1–2 days on a single GPU.

For edge deployment: export to ONNX → TensorRT engine. YOLOv8n in TensorRT FP16 on Jetson AGX Orin gives 150+ FPS at P99 latency < 8 ms—3 times faster than ONNX Runtime without TensorRT. On server A10G: 700+ FPS for YOLOv8n in TensorRT INT8.

How Does Fine-Tuning YOLO Help in Pattern Recognition?

Suppose you need to find micro-defects on a metal surface—a task with high resolution and class imbalance. We use YOLOv8m pretrained on COCO and fine-tune on 2000 proprietary images. Apply augmentations Mosaic, MixUp, random perspective. After 200 epochs, mAP 0.5 reaches 0.93. Key techniques:

• Focal loss for the objectness head—reduces contribution of easily classified examples.
• Class-balanced sampling—equalizes representation of rare classes.
• Test Time Augmentation (TTA)—increases recall by 5–7% through averaging over flips and scales.

Get a consultation on architecture selection for your task—contact us.

Segmentation: SAM, Mask R-CNN, and Instance Segmentation

SAM (Segment Anything Model) from Meta changed the approach to segmentation. SAM 2 works with video, supports object tracking across frames—for interactive object selection by point or bbox, it's the best out-of-the-box choice. For production instance segmentation without interactive prompting, Mask R-CNN or YOLOv8-seg are used. YOLOv8-seg trains like a regular detector with additional masks, convenient in the same pipelines. Semantic segmentation (each pixel is a class) uses SegFormer, DeepLabV3+. SegFormer-B5 provides a good balance of accuracy and speed for satellite imagery or medical segmentation.

Case study: cell segmentation on microscopic images. Dataset of 400 images with manual annotation. Training Mask R-CNN on ResNet-50 backbone gave IoU 0.61—poor. Problem: objects (cells) overlap; standard NMS kills overlapping predictions. Solution: switch to cellpose (specialized architecture for biomedical tasks) + soft-NMS. IoU increased to 0.79.

OCR: When Tesseract Fails

Tesseract is a starting point for simple tasks: printed text, good lighting, straight layout. As soon as there are handwritten elements, non-standard fonts, perspective distortions, or multi-column layouts, Tesseract degrades quickly.

PaddleOCR is a production-grade solution: text block detection + recognition + structural analysis. Works out of the box for 80+ languages, including Russian. Supports tables and complex document structures. TrOCR (Microsoft) is a transformer OCR with strong results on handwritten text. For Russian handwritten text, fine-tuning is needed: the base model is trained mostly on Latin script.

What to Do When Tesseract Cannot Handle Pattern Recognition on Documents?

For tasks like "extract data from invoices/contracts/passports," we use LayoutLMv3 or Donut—these models understand document layout, not just text. Integration via Hugging Face Transformers, fine-tuning on 200–500 annotated documents. Typical pipeline:

1. Preprocessing: deskew, denoising, binarization via OpenCV.
2. Text block detection: PaddleOCR detection or CRAFT.
3. Recognition: PaddleOCR recognition or TrOCR.
4. Post-processing: normalization, validation via regex or LLM for structured fields.

For documents with fixed structure, template matching + OCR by coordinates is often more reliable than an end-to-end solution.

Face Recognition: Identification and Verification

Face recognition = detection + alignment + embedding + matching. Each stage matters.

Detection: RetinaFace or InsightFace for accurate face localization and keypoints. MTCNN is older but reliable. Embedding: ArcFace (InsightFace) is state-of-the-art for face recognition embeddings. Models iresnet50/iresnet100 pretrained on MS1MV3 (5M identities). Embedding vector 512 float32, comparison by cosine similarity. Threshold tuning: decision threshold is a critical parameter. At threshold 0.6, typical FPR on LFW benchmark is 0.001, TPR is 0.985. In production, threshold must be calibrated to the real distribution: people in masks, with changed appearance, different lighting conditions. Liveness detection is mandatory: MiniFASNet—lightweight model on CPU; FaceX-Zoo contains several pretrained liveness detectors.

Video Analytics

Video is a sequence of frames plus a temporal dimension. A naive approach—detecting on every frame—is expensive.

Tracking: ByteTrack and BoT-SORT are the standard for multi-object tracking. They work on top of any detector, adding persistent IDs to objects across frames—enabling object counting, motion tracking, velocity.

Optimization: not every frame needs processing. For static scenes, detect every 5–10 frames, with tracking in between. For event detection (person entering a zone), background subtraction (OpenCV MOG2) serves as a lightweight pre-filter before neural detection. Action recognition: SlowFast, VideoMAE for action classification. Heavy models—for production use ONNX export + TensorRT or offline processing.

How to Measure Pattern Recognition Model Quality in Production?

Quality monitoring is key to MLOps. We track:

• Prediction confidence distribution.
• Share of low-confidence predictions (indicator of OOD data).
• Drift of input images via feature distribution (embeddings from backbone).

A drop in average confidence from 0.87 to 0.71 over a week is an early signal of distribution shift. NVIDIA Triton Inference Server recommends tracking these metrics via Prometheus. Our certified engineers set up monitoring and guarantee SLA for inference quality.

Deployment of CV Models

For online inference, we use Triton Inference Server (NVIDIA)—production standard for serving CV models. Supports TensorRT, ONNX, PyTorch, dynamic batching, multiple instances. REST and gRPC API. We guarantee stable operation under load.

Edge deployment: ONNX Runtime on ARM/x86 CPU. TensorFlow Lite for mobile devices. OpenVINO for Intel CPU/GPU/VPU—gives 2–3× speedup on Intel hardware compared to ONNX Runtime. After deployment, we hand over the model with documentation and train personnel.

What Is Included in the Work

Deadlines and Cost

A prototype detector on existing data takes 1–2 weeks. Production system with optimization for target hardware takes 4–8 weeks. Full cycle including data labeling (1000–5000 images) takes 2–4 months. Cost is calculated individually for each task. Typical savings from implementing a quality control system can be significant per production line.

We have been in the market for over 5 years and completed 60+ computer vision projects. We will evaluate your project end-to-end—request a consultation to get a quote and technical proposal.