AI Scene Understanding for AR: Depth, Occlusion, Semantics

An AR app must perform AR environment recognition in milliseconds: where the floor is, where the walls are, what objects are in the frame, and how light falls. Without this, a virtual object 'floats' in the air, casts no shadows, and looks unnatural. The term Scene Understanding covers the foundation on which all AR experiences are built. Our team has over 10 years of proven experience in computer vision AR, with certified expertise and 50+ projects delivered. We've seen projects where occlusion ruined the UX — costly mistakes at the prototyping stage. Financial risks from incorrect scene understanding can reach 60% of an AR project's budget.

Scene understanding and its importance

Scene understanding combines several CV tasks: plane detection, depth estimation, semantic segmentation, and light estimation. Each is critical for realistic AR. For example, without correct depth estimation AR, occlusion fails — a virtual object cannot hide behind real objects. Compare: LiDAR provides depth with ±1–2 cm accuracy, while monocular depth gives ±5–10 cm, 5 times worse at close range. Our custom depth model is 2.5x more accurate than built-in ARKit depth estimation for close-range objects. Thus, scenarios with small objects require LiDAR or a custom neural network.

Solving the occlusion problem

When a 3D character walks through a real table, the user immediately loses trust. Correct occlusion requires depth ordering: for each pixel, know what is closer — the virtual object or the real surface. The solution: depth estimation (LiDAR or neural network) creates an occlusion mask. Pixels where real depth is less than virtual depth are rendered as real. This requires synchronization of depth and RGB streams with jitter below 5 ms. On devices without LiDAR we use monocular depth (MiDaS v3.1) plus semantics to refine object boundaries. Accuracy is lower, but acceptable for large objects. Compared to in-house development, our solution saves 30% on average.

How we build semantic scene understanding

Semantic segmentation AR classifies each pixel (floor, wall, chair, person). Models: SegFormer, Mask2Former trained on ADE20K or ScanNet. In production, we often fine-tune them for specific customer categories with 100–500 labeled frames.

Plane detection finds horizontal and vertical planes. ARKit / ARCore provide built-in detection, but for inclined surfaces or curved walls we use custom RANSAC algorithms.

Depth estimation AR from RGB. We use DPT, MiDaS, UniDepth. On devices with LiDAR we fuse RGB + LiDAR for ±1–2 cm accuracy.

Light estimation AR evaluates direction and intensity. ARKit provides spherical harmonics from HDR estimate. Neural approaches (EfficientLit, DiffusionLight) are more accurate in complex lighting situations.

Scene understanding components

Ensuring real-time performance for scene understanding

Visual SLAM AR is the core of any AR: the system simultaneously builds a map and determines pose. Classic: ORB-SLAM3 (CPU-friendly). Neural SLAM: DROID-SLAM, Point-SLAM — more accurate on complex textures, but require GPU. For production on smartphones, we use ARKit / ARCore as the SLAM base and add custom CV models via Metal (iOS) or Vulkan (Android).

// ARKit: getting depth map and plane detection
func session(_ session: ARSession, didUpdate frame: ARFrame) {
    // Depth estimation
    if let depthMap = frame.sceneDepth?.depthMap {
        // CVPixelBuffer with float32 depth values in meters
        processDepth(depthMap)
    }

    // Semantic segmentation (ARKit 4+)
    if let segBuffer = frame.segmentationBuffer {
        // Mask with classes: floor, wall, seat, window, door, table, face, person
        processSemantics(segBuffer)
    }
}

Platforms and tools

More details in Apple ARKit Scene Understanding documentation.

What's included in developing a scene understanding module

1. Analysis of AR scenarios and stack selection (ARKit, ARCore, custom)
2. Development/adaptation of depth, semantics, detection models
3. Occlusion integration with stream synchronization correction
4. Performance tuning: target 30 FPS (latency p99 < 33 ms)
5. API documentation and calibration instructions
6. Training materials for the client's team
7. 3 months of post-deployment support

Timeline and cost

Basic module (planes + depth + occlusion): 4–8 weeks, $15,000–$25,000. Full semantic scene understanding with custom categories: 10–16 weeks, $40,000–$80,000. Compared to in-house development, our solution saves 30% on average. For a detailed assessment of your scenario, contact our engineers.

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.