<<<<Previous Blog Next Blog>>>>
Audio & Visual Cue System for Anosmia (Smell Disorder) and Smart WheelChair - Ft. Amazon's Alexa
Week 11: Sep 11 - 14
It is the last week of the 'Design for a Cause' design challenge. After two months of fun, long hours of work, editing, programming, asking Alexa weird questions, the project has finally come to an end. This blog will cover the machine learning process of identifying smell with a demo. Also, this will include a video demo of the 'Smart Wheelchair ' part of the project in order to provide an overall look at the project in total. For the introductory post about the 'Cue System for Anosmia' part of my project, visit this link - Cue System for Anosmia and Smart WheelChair #10 - Gas Sensors and Machine Learning.
Quick Links
1. Project Blogs Collection - Design for a Cause - Design Challenge
2. Plan - Cue System for Anosmia and Smart WheelChair #3 - The Plan
3. Design Challenge Page - Design for a Cause - Design Challenge
4. Related Discussion - Is it possible to transmit and receive smell?
Demo of 'Cue System for Anosmia'
Below is the demo video of the completed 'Cue System for Anosmia' part of my project.
Machine Learning the Smell
For machine learning the smell, I have used the AWS machine learning service. This will create a machine learning model from the data source we provide, which has to be placed in the AWS S3 storage service. Once the model is created, we can try real-time predictions of smell in the AWS machine learning console. Also, we can create an endpoint using which we can request and receive predicted smell information in a JSON format. I have created a web service in my local which will make a request to the Machine learning endpoint with data inputs from the mq5 and the mq2 gas sensor connected to Arduino MKR1000. This will be used by the machine learning model to predict and provide a JSON response. From this response, the smell information will be extracted. All this happens, when we ask "Alexa, how does that smell like". The output will be passed to Alexa to voice out, which will be the name of the smell. Also, Alexa can check the environment when asked to do so (not automatically) and alert if there is a dangerous smell in the environment. For now, I have programmed the mosquito coil as the dangerous smell. When this is sensed, Alexa will voice out the alert message and then Arduino MKR1000 will turn ON a fan to probably extinguish the mosquito coil (smoke)! It can also turn OFF an electric stove if it is connected to the extension box which is in turn connected to the relay.
The AWS machine learning costs as per the usage. The other services used in this project are free of cost. To provide an overview of how much it costs for usage, a screenshot of my bill for initial usage is given below.
Since this is neither completely an open source service or was I able to find one within this timeline, I will provide the documentation which I followed to create the machine learning model. Here is the documentation from Amazon for setting up ML - Click here, for a tutorial - Click here and for an introduction to AWS ML video from YouTube - Click here (this is not from Amazon). For the dataset/ data source which I used to create an ML model - Click here.
The machine learning model uses a multiclass type of prediction and below is a screenshot of a real-time prediction of smell in AWS machine learning console.
Demo of 'Smart WheelChair' [Repost]
Though this has been already posted in this blog - Cue System for Anosmia and Smart WheelChair #9 - Smart WheelChair Integration, I am reposting it again so that this blog has the overall summary and demos of all the parts of the project. Below is the demo video of the Smart WheelChair part of my project.
Summary
It has been fun over the two months building this project and making it a reality. The future scope/ plan would be delivering this as a product to the public in large scale. But I don't have a clue on how and I will post a blog here if that happens. The learning I had is immense, be it from my project or from the projects of other contestants. Arduino MKR1000 has been a wonderful, efficient and easy IoT device alongside the ESP8266 device family. Alexa, with this programmable capability make it a best option for voice command based automation/ data retrieval projects. IFTTT and thinger.io were also the best platforms I have seen so far to use for IoT projects. Enough of me talking, let's ask Alexa, how the experience was, in the video below.
Table of Contents/ Project Index
Have you got any suggestion or comment? Let me know in the comments section below.
Progress made so far,
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||| 100%
<<<<Previous Blog Next Blog>>>>
Top Comments
Great conclusion!! It has been a quick 2 months and now it's almost over. Nice to see your project complete, congratulations!!
Dale
Congratulations on finishing your project dixonselvan, I have enjoyed following it along the way. Good luck in the competition.
Kind regards
Thank you for following along three-phase your comments encouraged me in this 2 month journey.