Scribd
Upload
29 views9 pages

Object Detection History 1707305921

The history of object detection in computer vision began in the 1970s with early work laying the foundation for techniques like the Viola-Jones object detector released in 2001. In 2005, Histogram of Oriented Gradients (HOG) features became popular for object detection. The breakthrough of deep learning with convolutional neural networks (CNNs) revolutionized object detection, beginning with Region-based CNN (R-CNN) in 2014 and subsequent models like Fast R-CNN and Faster R-CNN that improved speed. You Only Look Once (YOLO) was introduced in 2016 for real-time object detection using a single-shot approach.

Uploaded by

singhebsl
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
29 views9 pages

Object Detection History 1707305921

The history of object detection in computer vision began in the 1970s with early work laying the foundation for techniques like the Viola-Jones object detector released in 2001. In 2005, Histogram of Oriented Gradients (HOG) features became popular for object detection. The breakthrough of deep learning with convolutional neural networks (CNNs) revolutionized object detection, beginning with Region-based CNN (R-CNN) in 2014 and subsequent models like Fast R-CNN and Faster R-CNN that improved speed. You Only Look Once (YOLO) was introduced in 2016 for real-time object detection using a single-shot approach.

Uploaded by

singhebsl
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Computer Vision

HISTORY OF OBJECT DETECTION

A R U N A R U N I S T O
The roots of object detection research can
indeed be traced back to the 1970s. Early
work in computer vision and image
processing laid the foundation for object
detection techniques.
The Image Understanding Environment
(IUE) project(1970's), led by researchers
such as David Marr and others, aimed to
develop computer algorithms for
understanding the contents of images.
The first real object detector was viola jones
developed in 2001.

The Viola-Jones object detection framework,


initially designed for face detection, gained
popularity.
It utilized Haar-like features and a cascaded
structure to efficiently detect objects. While
successful for faces, it had limitations in
handling diverse object categories.

The other method is Histogram of Oriented


Gradients(HOG) was released in 2005.
HOG features became popular for object
detection, especially for pedestrian
detection. It focused on the distribution of
gradient orientations in images, providing
robustness to variations in appearance.
Then came the revolution of AI with neural
networks, the breakthrough of deep
learning, particularly convolutional neural
networks (CNNs), revolutionized object
detection.

The ImageNet Large Scale Visual


Recognition Challenge (ILSVRC) played a
crucial role by providing large datasets for
training. AlexNet (2012) and subsequent
CNN architectures significantly improved
object recognition accuracy.
But this solves classification problem but
not detection problem.
To solve this issue where Region-based CNN
(R-CNN) was developed in 2014.
R-CNN introduced the idea of using region
proposals to localize and classify objects. It
utilized selective search for region proposals
and a CNN for feature extraction.
R-CNN gave good results but it was very
slow, then came the Fast-RCNN and Faster-
RCNN.
Fast R-CNN (2015) improved upon R-CNN by
sharing convolutional layers for region
proposals and classification. This led to
faster inference times and better
performance.

Faster R-CNN (2015) integrated region


proposal networks (RPNs) directly into the
object detection pipeline. The introduction
of RPNs eliminated the need for separate
region proposal methods, improving speed.
Faster-RCNN
Fast RCNN and Faster RCNN is faster than
RCNN but still not real-time.

Then there is breakthrough after the YOLO


introduced in 2016.
You Only Look Once YOLO (2016) were
designed for real-time object detection.
YOLO adopted a single-shot approach,
predicting bounding box coordinates and
class probabilities directly from the entire
image.

RCNN works with multiple iterations but


yolo done it by single iteration, this makes
yolo more efficient and faster.
Architecture of the YOLO is based on the
convolutional layers similar to the
predecessors.

YOLO has been widely adopted in various


applications, including object tracking,
autonomous vehicles, surveillance, and
robotics. Its ability to provide real-time
object detection with competitive accuracy
has contributed to its popularity in the
computer vision community.

You might also like