image

YOLO DETEC.

Real time object detection using a YOLO machine learning framework.

DEMO

What
It is

YOLO is a convolutional neural network based model that detects objects in real time using the "You Only Look Once" framework. It is based in darkfflow and can detect over 9000 different objects with 70% accuracy.

The algorithm runs up to 60fps, 12x faster than competing model Faster R-CNN.











How it
Works

Each frame of live footage is inputted directly into the algorithm at a rate of 60fps.

The YOLO framework applies a convolution layer to the frame, reducing its size to a 13x13 matrix.

Each cell in the matrix predicts 5 bounding boxes, each associated with one of the 9000 classes.

Binding boxes with a confidence score of >30% are shown to the user with their respective class label.





1.

INPUT

2.

REDUCTION

3.

PREDICTION

4.

THRESHOLDING





Convolutional Neural Network


YOLO uses a CNN framework for object detection.

CNN
Analysis

1.
Convolution

A weighted kernel of size NxN is slid over the image, generating a feature map as it multiplies pixel values by kernel values.

2.
Max Pooling

A seperate kernel of size NxN slides over the featuure map, passing on only the largest pixel values to a vector.

3.
Prediction

Values in the max pooling vector are multiplied by weights, and the resulting values are associated with their respective classes.

Phone :

647 236 4460

Email :

lukewpiette@gmail.com

Linkedin :
Luke Piette