Automate UI without API: CV-Powered RPA Bots Turnkey

Imagine: a legacy COBOL system, no API, and you need to automate data entry. A classic RPA (UiPath, Automation Anywhere) relying on XPath will break at the first button shift. The alternative — an RPA bot with Computer Vision that 'sees' the screen as an image and finds elements by visual features. We develop such bots turnkey, leveraging YOLOv8, PaddleOCR, and few-shot learning methods, delivering intelligent automation. Over 5+ years, we've completed 20+ CV-RPA projects, including complex cases with Citrix and dynamic UIs. We guarantee recognition stability: detection accuracy of 95%+ with a proper dataset.

For element detection we use YOLOv8, fine-tuned on the Rico dataset (66k UI) plus custom annotation for the client's interface. Augmentation (brightness, contrast, rotation) and INT8 quantization reduce latency to 50 ms. Text is extracted with PaddleOCR (Cyrillic) or EasyOCR fine-tuned on a domain dictionary. As a result, the bot processes up to 10 operations per minute.

According to Ultralytics documentation, YOLOv8 achieves mAP 50-95 on UI datasets, confirming its applicability for interface automation.

When Is CV-RPA Needed?

Computer vision for RPA is justified in specific scenarios:

• Working with legacy systems without API and with a closed window hierarchy (COBOL/AS400 terminals, Citrix Virtual Desktop).
• Web applications with dynamically generated class names (React/Angular with CSS Modules), where XPath is unstable with every deployment.
• Processing PDF documents and scanned images within an RPA flow — without OCR, the integration is incomplete.
• Automating third-party desktop applications without SDK — SAP GUI or 1C in terminal mode.

CV-RPA Bot Architecture

import cv2
import numpy as np
from ultralytics import YOLO

class CVRPAAgent:
    def __init__(self, ui_detector_model: str):
        self.detector = YOLO(ui_detector_model)
        self.screenshot_engine = ScreenshotEngine()

    def find_element(self, element_type: str,
                     text_hint: str = None) -> tuple[int, int]:
        screenshot = self.screenshot_engine.capture()
        detections = self.detector.predict(screenshot, conf=0.7)

        candidates = [d for d in detections if d.class_name == element_type]
        if text_hint:
            candidates = self._filter_by_ocr_text(candidates, screenshot, text_hint)

        if not candidates:
            raise ElementNotFoundError(f"Cannot find {element_type}")

        best = max(candidates, key=lambda d: d.confidence)
        return best.center_x, best.center_y

    def click(self, element_type: str, text_hint: str = None):
        x, y = self.find_element(element_type, text_hint)
        pyautogui.click(x, y)

For UI element detection, we use YOLOv8 (GitHub), fine-tuned on a dataset of UI components (buttons, input fields, checkboxes, dropdowns). The base model is the Rico Dataset (66k Android UI) plus custom annotation for the specific client interface. During training, we apply augmentation (brightness change, contrast, rotation) for robustness to different screenshot conditions.

OCR Integration for Data Reading

For extracting textual data from the screen: PaddleOCR (best speed-accuracy balance for Cyrillic) or EasyOCR. Integration into the flow: find element → extract text from ROI (Region of Interest) → pass to processing logic.

import paddleocr

ocr = paddleocr.PaddleOCR(use_angle_cls=True, lang='ru')

def extract_text_from_region(image, bbox):
    x1, y1, x2, y2 = bbox
    region = image[y1:y2, x1:x2]
    result = ocr.ocr(region, cls=True)
    return ' '.join([line[1][0] for line in result[0]])

How CV-RPA Handles Dynamic UI

Dynamic elements (pop-ups, changing button positions) are a typical challenge. We solve this with few-shot learning: collect 5–10 screenshot variants with different element states and fine-tune the model on them. This provides resilience to changes without full retraining. Additionally, we use grid search on the confidence threshold — tuning the conf threshold to minimize false positives.

Why CV-RPA Is More Reliable Than Classic RPA

CV-RPA is 3 times more resilient to UI changes compared to classic XPath-based RPA. Here's a comparison of key metrics:

How to Implement CV-RPA: Step-by-Step Process

1. Analysis and data collection — screenshots of the target UI, bounding box annotation for each element type.
2. Detector training — fine-tuning YOLOv8 on the collected dataset, validation on a test set.
3. Integration with RPA platform — embedding the CV agent into UiPath / Automation Anywhere via Python Activity.
4. Testing on real scenarios — running 100+ operations, measuring accuracy and time.
5. Deployment and monitoring — installation on the RPA farm, error logging, cyclic fine-tuning.

What's Included in the Work

• Fine-tuned YOLOv8 model for your UI
• OCR integration (PaddleOCR for Cyrillic)
• Architecture and fine-tuning documentation
• 3 months of support after delivery
• Source code of the agent and deployment configs

Timeline and Cost

Pricing is tailored to each project. Contact us for a free consultation — we'll discuss your task and choose the optimal solution. Order a pilot project: test CV-RPA on one process, evaluate accuracy and execution time.

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.