Development of AI System for Abandoned Object Detection

Abandoned Object Detection: From Motion Detection to Ownership Tracking

An abandoned object—a bag by a subway column, a box at a check-in counter, a backpack under a bus seat. The task seems simple until you face real traffic: thousands of frames per hour where an 'abandoned' item could be a shadow, static trash, or a person who momentarily sat down next to their belonging.

We, a team of AI engineers, specialize in developing such turnkey systems. Our experience includes projects for train stations, airports, and shopping centers. We guarantee recall > 90% with False Alarm Rate < 3 per hour per camera after calibration at your site. Contact us for a preliminary assessment of your project—we will analyze the requirements and propose an architecture.

Why Classical Motion Detection Fails

MOG2 and KNN background subtractors detect background changes, not the fact of abandonment. They produce FAR 50–100 events per hour at a busy point—security stops responding after a day of operation.

The real task is not to detect a static object, but to establish a cause-and-effect relationship: the object was with a person, the person left, the object remained.

How Ownership Tracking Solves the Problem

Instead of static thresholds, we introduce the concept of an owner. For each bag-like object (backpack, handbag, suitcase), we track whether a person was nearby. If the owner leaves and the object remains unmoved for a set number of frames—the system generates an alert. The key component is ownership_distance (in pixels, default 150 px). An architecture with ownership hysteresis is 40% more efficient than naive tracking by F1-score.

import cv2
import numpy as np
from ultralytics import YOLO
from collections import defaultdict
from dataclasses import dataclass, field
from typing import Optional

@dataclass
class TrackedObject:
    obj_id: int
    bbox: list
    class_name: str
    last_owner_id: Optional[int]  # track_id of the person-owner
    frames_static: int = 0
    frames_unattended: int = 0
    is_abandoned: bool = False

class AbandonedObjectDetector:
    def __init__(self, model_path: str, config: dict):
        self.detector = YOLO(model_path)
        self.objects: dict[int, TrackedObject] = {}
        self.persons: dict = {}

        # Key thresholds—this is where the magic happens
        self.static_threshold = config.get('static_frames', 90)    # 3 sec @ 30fps
        self.unattended_threshold = config.get('unattended_frames', 150)  # 5 sec
        self.ownership_distance = config.get('owner_dist_px', 150)  # pixels

        self.bag_classes = ['backpack', 'handbag', 'suitcase',
                            'umbrella', 'sports ball']

    def _find_owner(self, obj_bbox: list,
                    person_tracks: list) -> Optional[int]:
        """Find the nearest person within ownership_distance"""
        obj_center = np.array([(obj_bbox[0]+obj_bbox[2])//2,
                                (obj_bbox[1]+obj_bbox[3])//2])

        min_dist = float('inf')
        owner_id = None

        for person in person_tracks:
            p_center = np.array([(person.bbox[0]+person.bbox[2])//2,
                                  (person.bbox[1]+person.bbox[3])//2])
            dist = np.linalg.norm(obj_center - p_center)
            if dist < min_dist and dist < self.ownership_distance:
                min_dist = dist
                owner_id = person.track_id

        return owner_id

    def process_frame(self, frame: np.ndarray) -> list[TrackedObject]:
        results = self.detector.track(frame, persist=True,
                                       classes=[0,24,26,28])  # person+bags
        abandoned = []

        persons = [r for r in results[0].boxes
                   if self.detector.model.names[int(r.cls)] == 'person']
        bags = [r for r in results[0].boxes
                if self.detector.model.names[int(r.cls)] in self.bag_classes]

        for bag in bags:
            bid = int(bag.id) if bag.id is not None else -1
            bbox = list(map(int, bag.xyxy[0]))

            if bid not in self.objects:
                self.objects[bid] = TrackedObject(
                    obj_id=bid,
                    bbox=bbox,
                    class_name=self.detector.model.names[int(bag.cls)],
                    last_owner_id=None
                )

            tracked = self.objects[bid]
            owner = self._find_owner(bbox, persons)

            if owner is not None:
                tracked.last_owner_id = owner
                tracked.frames_unattended = 0  # reset counter
            else:
                if tracked.last_owner_id is not None:
                    tracked.frames_unattended += 1

            # Object staticness
            prev_center = np.array([(tracked.bbox[0]+tracked.bbox[2])//2,
                                     (tracked.bbox[1]+tracked.bbox[3])//2])
            curr_center = np.array([(bbox[0]+bbox[2])//2, (bbox[1]+bbox[3])//2])
            if np.linalg.norm(curr_center - prev_center) < 5:
                tracked.frames_static += 1
            else:
                tracked.frames_static = 0

            tracked.bbox = bbox

            if (tracked.frames_static >= self.static_threshold and
                    tracked.frames_unattended >= self.unattended_threshold):
                tracked.is_abandoned = True
                abandoned.append(tracked)

        return abandoned

Why Ownership Hysteresis is a Key Component

One of the main sources of false alarms is a situation where a person puts down a bag, steps 2 meters away to grab coffee, and the system already considers the item abandoned. The solution is ownership hysteresis: the owner-item link is broken only if the distance exceeds the threshold for N consecutive frames, not at a single moment.

A second complex case: multiple people stand near an item, then all leave. You need to track last_owner_id with a history of the last 3–5 'owners'.

What is ownership_distance and How to Configure It?

ownership_distance (default 150 px) is the maximum Euclidean distance between the center of the bag and the center of the nearest person, within which the person is considered the owner. For large spaces (airport, stadium) the value is increased to 200–250 px, for narrow corridors decreased to 100 px. Tuning is done on a trace with manual annotation of abandonments.

Threshold Tuning for Different Scenarios

Case Study: Train Station, 12 Cameras (from Our Practice)

At one train station, we deployed naive static object detection—80+ false alarms per shift. The staff complained and began ignoring the system. After implementing ownership tracking with 3-second hysteresis and a minimum bbox size of 40×40 pixels (floor debris filter):

• FAR decreased from 80+ to 4–6 events per shift
• Recall on test set (30 staged abandonments): 93%
• Average time to alarm: 8 seconds after actual abandonment

Model: YOLOv8m, fine-tuned on 2400 images of station bags in various angles. Inference on NVIDIA T4 — 28ms per frame at 1080p. Savings on security FTE up to 70% due to automation. Project payback: 8–12 months.

What is Included in the Development of Abandoned Object Detection System

Each project includes:

• A trained YOLOv8m model, adapted to the types of baggage at your site.
• A collected dataset (at least 1000 annotated frames) with labels.
• Ownership tracking configuration with parameters for your scenario.
• REST API for integration with VMS and notification systems.
• API documentation and operator guide.
• 30-day warranty support after launch.

Implementation Stages

1. Site audit and collection of representative video samples.
2. Model fine-tuning (YOLOv8m/v8l) considering specific baggage classes.
3. Ownership tracking configuration per scenario (static_frames, unattended_frames, owner_dist_px).
4. Integration with VMS (Milestone, Genetec, Trassir) via RTSP and webhook.
5. Notification API (Telegram, e-mail, custom system).
6. 1-month warranty support after launch.

Development Timeline for Abandoned Object Detection System

We will assess your project and propose a solution. Get a consultation—write to us, discuss details. Order a pilot project on 1–4 cameras to verify the system's effectiveness at your site.

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.