1 of 34

Curriculum for �Second Year B.Tech � Information Technology ��BIT24MD02 : Augumented Reality A.Y 2025-26�Sem – II

By-

Alpana A Borse

Asst. Prof.

IT Department

PCCOE, Pune

Introduction to Augmented Reality (A.R.)

1

2 of 34

Unit IV: AR Techniques- Marker based & Markerless tracking (8Hrs)

Introduction to Augmented Reality (A.R.)

2

3 of 34

Syllabus

PCCOE, Pune Introduction to Augmented Reality (A.R.)

3

    • Marker-based approach- Introduction to marker-based tracking, types of markers, marker camera pose and identification, visual tracking, mathematical representation of matrix multiplication
    • Marker types- Template markers, 2D barcode markers, imperceptible
    • markers.
    • Marker-less approach- Localization based augmentation, real world examples
    • Tracking methods- Visual tracking, feature based tracking, hybrid tracking, and initialization and recovery.

4 of 34

Marker Tracking in Augmented Reality

Marker-based AR uses predefined visual patterns that a camera detects to determine position and orientation in 3D space — forming the foundation of many AR experiences.

PCCOE, PUNE

INTRODUCTION TO AUGMENTED REALITY

5 of 34

How Marker Tracking Works

The Detection Pipeline

01

Detect the Marker

Camera captures a predefined marker in the scene.

02

Analyze Shape & Position

Computer vision algorithms parse the marker's geometry.

03

Calculate Pose

The system estimates camera position and orientation.

04

Render Virtual Object

Digital content is superimposed on the marker in real time.

Key Idea: Camera detects marker → calculates pose → renders virtual object

6 of 34

Marker Tracking — Applications

Education

AR-enabled books and educational materials bring static content to life with interactive 3D overlays.

Marketing

AR business cards and interactive ads create memorable, engaging experiences for audiences.

Gaming

Tabletop AR games use physical marker cards to trigger and control virtual game elements.

7 of 34

Taxonomy of AR Tracking Techniques

Two Major Branches

AR tracking techniques are broadly divided into two families:

Marker-Based

Uses fiducial markers or tags (direct/indirect, RFID, barcode).

Markerless

Uses sensor data or computer vision — model-based, Visual SLAM, SFM, and more.

8 of 34

Marker Based

PCCOE, Pune Introduction to Augmented Reality (A.R.)

8

Categorization of augmented reality tracking techniques.

9 of 34

Marker Based

PCCOE, Pune Introduction to Augmented Reality (A.R.)

9

Categorization of augmented reality tracking techniques.

10 of 34

Marker less

PCCOE, Pune Introduction to Augmented Reality (A.R.)

10

Categorization of augmented reality tracking techniques.

11 of 34

Marker less

PCCOE, Pune Introduction to Augmented Reality (A.R.)

11

Categorization of augmented reality tracking techniques.

12 of 34

Marker less

12

13 of 34

13

14 of 34

Difference Between Marker-Based and Markerless Tracking

PCCOE, Pune Introduction to Augmented Reality (A.R.)

14

Feature

Marker-Based Tracking

Markerless Tracking

Definition

Uses predefined physical markers (QR codes, AR markers, fiducial markers) for tracking.

Uses computer vision and AI to recognize objects, features, or environments without predefined markers.

Tracking Method

Camera detects the position and orientation of the marker to place virtual objects.

Analyzes the environment using feature detection, SLAM (Simultaneous Localization and Mapping), and AI-based recognition.

Examples

- QR codes in AR apps

15 of 34

Marker-Based vs. Markerless Tracking

Choosing the Right Approach

Marker-based tracking excels in controlled environments where high accuracy is paramount. Markerless tracking opens the door to broader, real-world applications but demands significantly more computational resources and sophisticated algorithms.

16 of 34

Summary of Tracking Methods

PCCOE, Pune Introduction to Augmented Reality (A.R.)

16

Tracking Method

Key Features

Best Use Cases

Natural Feature Tracking (NFT)

Uses real-world textures instead of markers

Product packaging, museum exhibits, manuals

SLAM

Creates a dynamic 3D map of the environment

AR navigation, gaming, remote assistance

Outdoor Tracking

Uses GPS, IMU, and vision for large-scale AR

AR tourism, outdoor gaming, city navigation

17 of 34

Feature

Natural Feature Tracking (NFT)

SLAM

Outdoor Tracking

Purpose

Track features in images

Build map + track position

Track in real outdoor environment

Input

Camera images

Camera / LiDAR / sensors

GPS + sensors + camera

Main Work

Detect & follow points

Localization + Mapping

Large-scale tracking

Map Creation

❌ No

✅ Yes

✅ (sometimes via SLAM)

Position Tracking

❌ Limited

✅ Yes

✅ Yes

Environment

Small/local scenes

Indoor + outdoor

Mainly outdoor

Complexity

Low

High

Medium to High

18 of 34

Feature

Natural Feature Tracking (NFT)

SLAM

Outdoor Tracking

Purpose

Track features in images

Build map + track position

Track in real outdoor environment

Example

📱 Instagram filter tracks your face features (eyes, nose) and keeps the filter fixed while you move

🤖 Robot in a room creates a map of walls and furniture while moving and knows its position

🚗 Google Maps navigation tracks your location on roads using GPS while you travel

19 of 34

Types of AR Markers

Template Markers

Simple black-and-white patterns recognized via image matching. Example: square markers with unique internal designs.

2D Barcode Markers

Encoded markers like QR codes that store additional data (IDs, URLs). More robust and information-rich than template markers.

Imperceptible Markers

Invisible to the naked eye — use infrared or UV light. Preserve visual aesthetics while enabling precise tracking in advanced AR systems.

20 of 34

Camera Pose Estimation

What Is Camera Pose?

Camera pose describes exactly where the camera is in 3D space and how it is oriented relative to a detected marker.

Pose Estimation Steps

1

Detect Marker Corners

Locate the four corner points of the marker in the image.

2

Match Stored Pattern

Compare against the known marker template in the database.

3

Estimate Transformation

Calculate the spatial relationship between marker and camera.

4

Compute Pose Matrix

Output a transformation matrix used for rendering.

Position

x, y, z coordinates in 3D space

Orientation

Rotation along each axis

21 of 34

Mathematical Representation — Transformation Matrix

In AR, all spatial transformations are encoded as matrix multiplications. The complete transformation combines translation, rotation, and scaling into a single compact form.

Basic Formula

Transformation

Translation × Rotation × Scaling

The 4×4 Transformation Matrix T

Why a 4×4 Matrix?

Using homogeneous coordinates allows both rotation and translation to be expressed as a single matrix multiplication — essential for efficient real-time rendering in AR pipelines.

R (3×3)

Encodes rotation along x, y, and z axes

t (vector)

Encodes the 3D translation (position offset)

22 of 34

Final Rendering Equation

Once the pose matrix is computed, the virtual object is projected onto the screen using the standard graphics pipeline formula:

Projection Matrix

Converts 3D coordinates to 2D screen space using camera intrinsics (focal length, sensor size).

View Matrix

Represents the camera's pose — where it is and which way it is looking in the scene.

Model Matrix

Defines the virtual object's own position, rotation, and scale in world space.

23 of 34

Visual Tracking Techniques

Frame-by-Frame Tracking

Visual tracking continuously analyzes the live camera feed, updating the marker's position and orientation with every new frame to keep virtual objects locked in place.

Edge Detection

Identifies marker boundaries by finding sharp intensity changes in the image.

Corner Detection

Uses algorithms like Harris Corner Detector to find stable reference points on the marker.

Pattern Matching

Compares detected regions to stored marker templates for robust identification.

24 of 34

Marker-less Tracking in AR

Unlike marker-based approaches, marker-less tracking uses natural features and sensors to anchor digital content to the real world — no predefined targets required.

CHAPTER 2

"Track real-world features instead of artificial markers"

25 of 34

Localization-Based Augmentation

Position estimation techniques anchor AR content to real-world coordinates using a combination of hardware and algorithmic inputs.

GPS

Outdoor AR positioning via satellite-based geolocation

IMU Sensors

Accelerometer and gyroscope data for orientation and movement

SLAM

Simultaneous Localization and Mapping for dynamic environments

26 of 34

SLAM: Mapping and Tracking in Real Time

How SLAM Works

SLAM is the backbone of modern marker-less AR. It solves two interdependent problems simultaneously — without requiring any pre-existing map.

Map Building

Constructs a spatial model of the unknown environment in real time using visual and depth data

Camera Tracking

Continuously estimates the camera's position and orientation relative to the evolving map

27 of 34

Real-World Applications

Marker-less AR is already embedded in consumer and enterprise products across industries.

AR Navigation

Google Maps Live View overlays directional cues onto live camera feeds for pedestrian navigation

Furniture Placement

IKEA Place uses surface detection to preview true-to-scale furniture in your space before purchase

Mobile Gaming

Pokémon GO anchors virtual creatures to GPS coordinates and real-world surfaces

Industrial AR

Maintenance systems overlay repair instructions and diagnostics onto physical machinery

28 of 34

CHAPTER 3

Visual Tracking

Camera-Based Object Tracking

Visual tracking processes raw camera input using image analysis algorithms to follow objects or surfaces across frames — no markers needed.

Optical Flow

Tracks pixel motion between consecutive frames to infer movement direction and velocity

Template Matching

Compares regions of interest against a reference template to detect and follow known objects

Contour Detection

Identifies and tracks object boundaries using edge-detection algorithms

29 of 34

Feature-Based Tracking

Detects unique keypoints in a scene and matches them across frames to estimate camera motion and object pose.

SIFT

Scale-Invariant Feature Transform — robust to scale and rotation changes; computationally heavier

SURF

Speeded-Up Robust Features — faster approximation of SIFT using integral images and Hessian matrices

ORB

Oriented FAST and Rotated BRIEF — lightweight, real-time capable, and royalty-free; preferred for mobile AR

30 of 34

Feature-Based Tracking Pipeline

Estimate Motion

Compute pose via homography or PnP

Match Features

Find correspondences across frames

Extract Descriptors

Compute compact feature vectors

Detect Keypoints

Identify salient points in image

This four-stage pipeline is the core engine behind algorithms like ORB and SIFT — converting raw pixel data into reliable pose estimates for stable AR rendering.

31 of 34

Hybrid Tracking

Combines vision-based and sensor-based inputs for greater reliability across challenging real-world conditions.

Vision-Based

Camera and image processing for spatial understanding

Sensor-Based

IMU, depth, and LiDAR data for motion and orientation

Key Advantage

Maintains accuracy in low-light conditions and during occlusion

32 of 34

Initialization and Recovery

Initialization

The first step of any tracking session. The system identifies the starting position by detecting reliable features or surface planes and establishing an initial coordinate frame for AR content placement.

Poor initialization is the most common cause of unstable AR rendering — robust feature detection at startup is critical.

Tracking Recovery

Triggered when tracking is lost — due to fast motion, occlusion, or sudden lighting change. The system re-detects keypoints or re-localizes against the existing map to resume stable tracking.

SLAM-based systems can re-localize against a previously built map, enabling faster and more reliable recovery.

33 of 34

Marker-Based vs. Marker-less

Understanding the trade-offs helps teams choose the right tracking strategy for each deployment context.

Feature

Marker-Based

Marker-less

Requirement

Predefined printed marker

No marker needed

Accuracy

High (controlled environments)

Moderate (scene-dependent)

Flexibility

Low — tied to marker placement

High — works in open environments

Complexity

Simple to implement

Algorithmically complex

Use Case

Education, demos, product packaging

Real-world AR apps, navigation, gaming

For production-grade AR targeting uncontrolled environments, marker-less tracking is the preferred foundation — despite its higher implementation complexity.

34 of 34