By clicking “Accept”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts. View our Privacy Policy for more information.
September 5, 2022

What is Intersection over Union in object detection?

September 5, 2022

What is Intersection over Union in object detection?

What is object detection?

Object detection is a technique in computer vision used to identify as well as locate objects in digital images or videos. It detects a targeted object in the image or video.

What are the components of object detection?


It refers to the assignment of an object to an image. In other words, finding which classes of objects are present in an image or a video.


It seeks to locate the most visible object in an image or video. Usually, it is centered within a tightly cropped bounding box. It easily ascertains the location of an image object.

What is Intersection over Union(IOU)?

It describes the level of overlapping between two boxes, that is, the predicting box and the actual bounding box. The greater the overlap, the greater the IOU. 

In object detection, Intersection of Union is a unit of measurement for checking the level of overlap between an object's predicted box and its actual bounding box in a particular dataset. Its symbol is $IoU$.

Why do we use Intersection over Union?

The predicting box and the actual bounding box are usually most unlikely going to be the same. This is where Intersection over Union is needed. It measures the overlap between both boxes, and as such determines the level of accuracy of the object detector.

What are the Object Detection techniques?

Scale-Invariant Feature Transform (SIFT)

David Lowe is the originator of SIFT. It uses scale and rotation invariants to detect the local features of an image also known as its key points. Image size and orientation do not affect it. There are four main steps in the SIFT process.

(i)Scale space construction

This ensures the scale independence of features.

(ii) Localization of key point

This is where the features of the image are identified.

(iii) Orientation assignment

This is where we ensure that the invariance of key points are rotated.

(iv) Key point descriptor

This is where a unique fingerprint is assigned to every key point

Speeded Up Robust Feature (SURF)

SURF is a patented object detector first introduced by Herbert Bay at the 2006 European Conference on Computer Vision. It has three main areas:


It uses the blob detector, which bases on the Hessian matrix, to find interest points. The calculated Hessian matrix is the measure of change of the interest point and the maximum value of determinant points which are selected.


Here, it provides unique descriptions of an image feature. It fixes an orientation that is reproducible based on information received from the circular region around the point of interest.


Here we find pairs that match with the use of descriptor comparison accessed from different images.

Features from Accelerated Segment Test(FAST) Corner Detector

It is a method of corner detection with computational efficiency, introduced by Tom Drummond and Edward Rosten. It provides corner points which are used as key points. It is good for real-time video processing applications due to a high speed performance. However, it is unable to detect corners on images with coordinates which are computer-generated. It assigns a segment per pixel in a circle of 16 pixels around the pixel.

You Only Look Once(YOLO)

YOLO detects different image objects in real-time through the use of neutral networks. A single logarithm run in yellow is enough for image detection. 

How is iOu measured? 

The formula for calculating Intersection over Union is called The Jaccard Index. The area of intersection for the two boxes is calculated. Afterward it is divided by the area of the Union of the two boxes. That means the intersection area is divided by the union area.

IoU= (ANB)÷ (AUB) or (I) ÷(U)

How does object detection work?

Object detection draws a bounding box around a detected object, that allows the specific object location to be detected, or alternatively how they move in a particular scene.

The major challenge of intersection over Union in object detection 

Its accuracy measure is a bit questionable. This is because its formula does not really reflect whether the two bounding boxes are in proximity or far from one another. 

You might also like
this new related posts

Want to scale up your data labeling projects
and do it ethically? 

We have a wide range of solutions and tools that will help you train your algorithms. Click below to learn more!