Developing an Object Measurement System from Images or Video

This is a task we solve using computer vision. On a production line, a part just came out of the machine — hot (80°C). Contact measurement with a caliper deforms the surface, introducing an error of up to 0.5 mm. Non-contact measurement via camera: a single shot in 50 ms, accuracy ±0.1 mm, data straight to SCADA. No risk of burns or conveyor stoppage.

Problems We Solve

Quality control on the production line. The part just came out of the machine — deviations from the drawing are detected instantly. The system measures length, width, diameter, angles, and contour area in 50 ms. Out-of-tolerance triggers an alarm or line stop.

Field measurements without access to the object. Need to know the width of a crack in a wall or the diameter of a pipe at height? Place an ArUco marker next to it, take a photo — in a second the dimensions are known. Ladders and rulers are unnecessary.

Logistics and warehouse. Box dimensions on a conveyor belt are measured in motion. The system automatically assigns length, width, and height to each unit — optimizing container loading.

How to Ensure Millimetre-Accurate Measurements

The key parameter is scale. We use three methods:

1. Calibrated camera — photograph a calibration plate once from the working distance, compute pixels_per_mm. Accuracy ±0.1 mm. Ideal for static conveyors.

2. ArUco marker in the frame — a marker of known size (e.g., 50 mm) is placed next to the object. The system detects the marker, computes the scale, then measures the object. Accuracy ±1–5 mm, suitable for field conditions.

3. Stereo pair — two cameras spaced 10 cm apart produce a 3D point cloud. From this we extract overall dimensions, depth, volume. Accuracy ±0.5–2 mm on objects up to 1 m.

The choice of method depends on the task. For a conveyor with a fixed camera position — calibration. For a mobile tablet — ArUco. For large objects (furniture, pallets) — stereo or LiDAR+RGB.

How to Choose the Measurement Method?

• If the object moves on a conveyor at constant speed and the camera is fixed — choose a calibrated camera. Maximum accuracy ±0.1 mm and measurement time 50 ms.

• If you are in a warehouse or field — use an ArUco marker. Print a 5×5 cm marker, glue it onto a rigid base, and place it next to the object. The system will automatically determine the scale.

• If you need to measure volume or depth — use a stereo pair. Two cameras with a 10 cm baseline build a 3D model and output dimensions in millimetres.

Step-by-Step: How the System Measures an Object

1. Image capture. The camera takes a snapshot (or frame from video). Resolution and frame rate are configured.
2. Scale determination. The system finds the calibration object or ArUco marker, computes pixels_per_mm.
3. Object segmentation. We use thresholding (Otsu) or a neural network for complex contours.
4. Dimension extraction. From the contour we compute bounding box, minimal rectangle, perimeter, area.
5. Conversion to millimetres. Multiply pixel values by pixels_per_mm.
6. Result output. Send data as JSON, to a database, or to the operator's screen.

What the Work Includes

• Task analysis: Send us a photo/video of the object with shooting conditions (distance, lighting, motion). We choose the method and prepare an estimate in 1 day.
• Prototype: In 3–5 days we build a demo version for your scenario. You test it on real data.
• Development: We implement segmentation, calibration, and measurement algorithms. Stack: OpenCV, PyTorch (if a neural network for complex contours is needed), Docker, FastAPI.
• Integration: Embed the module into your PLC/SCADA, add REST API, MQTT, Modbus.
• Documentation and training: Deliver code, API description, operator instructions.
• Support: 3 months free support, then by contract.

Timeframes (Approximate)

Cost is calculated individually and depends on segmentation complexity, number of cameras, and need for neural networks. Write to us — we will give a preliminary estimate in 1 day.

Why Choose Us?

Over 10 years of experience in industrial CV. We have implemented 50+ projects for mechanical engineering, logistics, and medicine. Accuracy guarantee: we record metrological characteristics in the contract. We always provide a validation report. Development processes are standardized according to ISO 9001. Compared to Western Halcon/Matrox solutions, our solution is 2–3 times cheaper with comparable accuracy, and implementation time is 1.5 times shorter.

Our customers save up to 2 million rubles per year on quality control. For example, one client reduced inspection time by 80%, equivalent to savings of about 1.5 million rubles per year. Get a consultation from an engineer with 10+ years of CV experience — contact us.

Scaling Methods

Method 1: Calibrated Camera

Once we photograph an object of known size (calibration plate) from the working distance, compute the pixels_per_mm coefficient:

import cv2
import numpy as np

class CalibratedMeasurement:
    def __init__(self, pixels_per_mm: float,
                 camera_matrix: np.ndarray = None,
                 dist_coefficients: np.ndarray = None):
        self.ppm = pixels_per_mm
        self.camera_matrix = camera_matrix
        self.dist_coefficients = dist_coefficients

    def calibrate_from_reference(self, image: np.ndarray,
                                   known_width_mm: float) -> float:
        """Calibration using an object of known width"""
        # Assume object is already aligned and occupies ~80% of width
        gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
        _, binary = cv2.threshold(gray, 0, 255,
                                   cv2.THRESH_BINARY + cv2.THRESH_OTSU)

        contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL,
                                        cv2.CHAIN_APPROX_SIMPLE)
        main = max(contours, key=cv2.contourArea)
        x, y, w, h = cv2.boundingRect(main)

        self.ppm = w / known_width_mm
        return self.ppm

    def measure_contour(self, image: np.ndarray) -> dict:
        """Measure object's key parameters"""
        if self.camera_matrix is not None:
            # Correct lens distortion
            image = cv2.undistort(image, self.camera_matrix,
                                   self.dist_coefficients)

        gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
        _, binary = cv2.threshold(gray, 0, 255,
                                   cv2.THRESH_BINARY + cv2.THRESH_OTSU)

        contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL,
                                        cv2.CHAIN_APPROX_SIMPLE)
        if not contours:
            return {'measured': False}

        c = max(contours, key=cv2.contourArea)

        # Bounding rectangle
        x, y, w, h = cv2.boundingRect(c)

        # Minimal rotated rectangle
        rect = cv2.minAreaRect(c)
        (cx, cy), (rw, rh), angle = rect
        min_side = min(rw, rh)
        max_side = max(rw, rh)

        # Perimeter via arc length
        perimeter_px = cv2.arcLength(c, True)

        # Area
        area_px = cv2.contourArea(c)

        return {
            'width_mm': round(w / self.ppm, 3),
            'height_mm': round(h / self.ppm, 3),
            'length_mm': round(max_side / self.ppm, 3),
            'width_min_mm': round(min_side / self.ppm, 3),
            'angle_deg': round(angle, 2),
            'perimeter_mm': round(perimeter_px / self.ppm, 3),
            'area_mm2': round(area_px / self.ppm**2, 3),
            'center': (round(cx / self.ppm, 2), round(cy / self.ppm, 2))
        }

Method 2: Reference Object in Frame (ArUco Markers)

import cv2.aruco as aruco

def measure_with_aruco(image: np.ndarray,
                        marker_size_mm: float = 50.0) -> dict:
    """Measurement using ArUco marker as reference"""
    aruco_dict = aruco.getPredefinedDictionary(aruco.DICT_4X4_250)
    detector = aruco.ArucoDetector(aruco_dict)

    corners, ids, _ = detector.detectMarkers(image)

    if ids is None:
        return {'error': 'no_aruco_marker_found'}

    # Compute pixels_per_mm from the marker
    marker_corners = corners[0][0]
    marker_width_px = np.linalg.norm(marker_corners[0] - marker_corners[1])
    ppm = marker_width_px / marker_size_mm

    # Then standard measurement
    measurer = CalibratedMeasurement(pixels_per_mm=ppm)
    return measurer.measure_contour(image)

3D Measurement via Stereo Pair

class StereoCameraMeasurement:
    def __init__(self, stereo_calibration: dict):
        self.Q = stereo_calibration['Q']          # disparity-to-depth matrix
        self.baseline_mm = stereo_calibration['baseline_mm']
        self.focal_length_px = stereo_calibration['focal_length_px']

    def measure_3d(self, left_img: np.ndarray,
                    right_img: np.ndarray) -> dict:
        # Stereo matching
        stereo = cv2.StereoSGBM_create(
            minDisparity=0,
            numDisparities=96,
            blockSize=11,
            P1=8 * 3 * 11**2,
            P2=32 * 3 * 11**2,
            disp12MaxDiff=1,
            uniquenessRatio=10,
            speckleWindowSize=100,
            speckleRange=32
        )

        disparity = stereo.compute(
            cv2.cvtColor(left_img, cv2.COLOR_BGR2GRAY),
            cv2.cvtColor(right_img, cv2.COLOR_BGR2GRAY)
        ).astype(np.float32) / 16.0

        # Convert disparity to 3D point cloud
        points_3d = cv2.reprojectImageTo3D(disparity, self.Q)

        return self._extract_dimensions_3d(points_3d)

Accuracy and Applications

According to a NIST study, non-contact measurement reduces error by 40%. Ready to discuss your task. Write to us — we will evaluate your project in 1 day. Get a consultation from an engineer.

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.