Scribd
Upload
31 views7 pages

Embedded Machine Learning in Vision

This document provides an introduction to computer vision with embedded machine learning. It discusses using neural networks like convolutional neural networks on microcontrollers and embedded devices to perform tasks like image classification and object detection in pictures with applications in devices like cameras, robots and drones. The document describes common computer vision and deep learning concepts like pixels, normalization, feature extraction and classification.

Uploaded by

jamer estrada
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
31 views7 pages

Embedded Machine Learning in Vision

This document provides an introduction to computer vision with embedded machine learning. It discusses using neural networks like convolutional neural networks on microcontrollers and embedded devices to perform tasks like image classification and object detection in pictures with applications in devices like cameras, robots and drones. The document describes common computer vision and deep learning concepts like pixels, normalization, feature extraction and classification.

Uploaded by

jamer estrada
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

Computer Vision with

Embedded Machine Learning


Welcome to the Course

© 2021 EdgeImpulse, Inc.


person: 0.72 person: 0.99
car: 0.98 car: 0.83
car: 0.95 car: 0.87

© 2021 EdgeImpulse, Inc.


picture element (“pixel”)

© 2021 EdgeImpulse, Inc.


Dense Neural Network (DNN)
Input layer Example:
- Accelerometer: -20…20 m/s2
- Light sensor: 0...120,000 lux
Hidden layer Normalize input: [0, 255] → [0.0, 1.0]

Output layer
Softmax
P(class1) P(class2) P(class3)

© 2021 EdgeImpulse, Inc.


Model Cat

© 2021 EdgeImpulse, Inc.


Input (2D array)
28x28

2D Convolution

2D Max Pooling
Feature extraction
2D Convolution

2D Max Pooling

Flatten to Vector 1x98

Classifier (DNN)
Softmax
P(class1) P(class2) P(class3) © 2021 EdgeImpulse, Inc.
• Object 1
– Class: dog (0.92)
– Bounding box
• (x1, y1)
• (w1, h1)
• Object 2
Object – Class: toy (0.85)
Detection – Bounding box
Model • (x2, y2)
• (w2, h2)
• Object 3
– Class: ball (0.77)
– Bounding box
• (x3, y3)
• (w3, h3)

© 2021 EdgeImpulse, Inc.

You might also like