Module 3: Robotic Vision

Santumol C Puthiyaparampil
Assistant Professor
Ilahia College of Engineering

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 1 / 21
 Robotic Vision – Introduction

Robotic vision refers to the capability of a robot to acquire visual information

from its surroundings and interpret it in a meaningful manner.
It involves capturing images using cameras, processing these images using
computer algorithms, and extracting useful information for decision making.
Robotic vision enables robots to identify objects, understand scenes, and perform
tasks intelligently in real-world environments.
It plays a vital role in automation, inspection, navigation, and intelligent robotic
applications.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 2 / 21
 Subareas of Robotic Vision

Robotic vision is divided into several functional stages, each responsible for a
specific task in image understanding.
The major subareas of robotic vision are:
Sensing
Preprocessing
Segmentation
Description
Recognition
Interpretation
Feature Extraction
Each stage contributes to the overall performance of a machine vision system.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 3 / 21
 Sensing

Sensing is the first and most important stage of robotic vision.

In this stage, visual information from the environment is captured using cameras
and visual sensors.
The quality of sensing directly affects the accuracy of all subsequent processing
stages.
Sensing includes camera calibration, image acquisition, illumination control, and
imaging geometry.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 4 / 21
 Camera Calibration

Camera calibration is the process of correcting errors in images caused by camera

lens distortion and improper orientation.
These distortions result in displacement of image points from their ideal positions.
Calibration ensures accurate mapping between real-world coordinates and image
coordinates.
The pinhole camera model is commonly used as an ideal reference model for
calibration.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 5 / 21
 Image Acquisition

Image acquisition is the process of converting visual information into a digital

image.
Light reflected from objects is captured by the camera sensor and converted into
electrical signals.
These signals are sampled spatially and quantized in amplitude to produce a
digital image.
Image acquisition determines the resolution and quality of the captured image.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 6 / 21
 Image Digitization

A digital image is mathematically represented as a two-dimensional function

f (x, y ).
Spatial digitization of the image is known as image sampling.
Amplitude digitization of pixel values is known as gray-level quantization.
The final digital image is represented and stored as a matrix of intensity values.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 7 / 21
 Illumination

Illumination refers to the lighting conditions under which an image is captured.

Proper illumination improves image quality and reduces complexity in image
processing algorithms.
Good lighting enhances important features required for object detection and
recognition.
Common illumination techniques include diffuse lighting, back lighting, structured
lighting, and directional lighting.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 8 / 21
 Imaging Geometry

Imaging geometry explains the relationship between three-dimensional objects and

their two-dimensional image projections.
It is governed by the laws of optics and camera geometry.
Due to small aperture size, camera lenses are often approximated using the
pinhole camera model.
Image points are formed by projecting object points through a common center of
projection.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 9 / 21
 Image Preprocessing

Image preprocessing is performed to improve the quality of images before further

analysis.
It removes noise and unwanted distortions introduced during image acquisition.
Preprocessing enhances important image features required for segmentation and
recognition.
This stage simplifies the processing tasks in later stages of robotic vision.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 10 / 21
 Preprocessing Methodologies

Image preprocessing can be carried out using two main approaches.

Spatial domain methods operate directly on image pixels.
Frequency domain methods process images using Fourier transforms.
Both methods aim to reduce noise and enhance useful image information.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 11 / 21
 Preprocessing Techniques

Common preprocessing techniques include neighborhood averaging and median

filtering.
Image enhancement techniques improve contrast and brightness.
Edge detection identifies boundaries between objects.
Image thresholding separates objects based on intensity levels.
These operations prepare the image for effective segmentation.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 12 / 21
 Image Segmentation

Image segmentation is the process of dividing an image into meaningful regions or

objects.
It separates the foreground objects from the background.
Segmentation is essential for object recognition and scene analysis.
Accurate segmentation improves the reliability of recognition results.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 13 / 21
 Segmentation Principles

Segmentation algorithms are based on two fundamental principles.

Discontinuity-based methods detect sudden changes in intensity, such as edges.
Similarity-based methods group pixels with similar properties using thresholding or
region growing.
Both approaches are widely used in robotic vision applications.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 14 / 21
 Image Description

Image description extracts meaningful features from segmented objects.

These features describe the shape, texture, and structure of objects.
Descriptors should be invariant to size, position, and orientation.
Effective description helps in accurate object recognition.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 15 / 21
 Image Descriptors

Common boundary and region descriptors include chain codes and polygonal
approximation.
Fourier descriptors represent object boundaries in frequency domain.
Texture descriptors describe surface properties such as smoothness and regularity.
Moment invariants are used as they remain unchanged under rotation, scaling,
and translation.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 16 / 21
 Image Recognition

Image recognition is the process of assigning labels to segmented objects.

It identifies objects as belonging to known categories.
Recognition assumes that objects are captured under known viewing conditions.
This stage converts visual data into meaningful object identities.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 17 / 21
 Recognition Methods

Decision theoretic methods use mathematical decision rules for classification.

Structural methods represent objects as combinations of simpler primitives.
Shape number matching and string matching are common structural techniques.
These methods enable accurate recognition of objects.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 18 / 21
 Image Interpretation

Image interpretation is the highest-level stage of robotic vision.

It combines sensing, preprocessing, segmentation, description, and recognition.
Interpretation extracts meaningful information from complex scenes.
It is affected by illumination changes, viewing geometry, and occlusions.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 19 / 21
 Feature Extraction

Feature extraction identifies the most important characteristics of an image.

It reduces the amount of data while preserving useful information.
Extracted features are used for classification and recognition tasks.
This stage improves computational efficiency and accuracy.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 20 / 21
 Conclusion

Robotic vision enables robots to perceive and understand their environment.

It consists of multiple sequential processing stages.
Each stage plays a crucial role in accurate image understanding.
Robotic vision is essential for intelligent automation and robotics systems.

Santumol C Puthiyaparampil Assistant Professor Ilahia

ModuleCollege of Engineering
3: Robotic Vision 21 / 21