Project14 | Vision Thing: Beaglebone AI Your Vision Thing Project!

Hi, I am already working on improving agriculture using wireless sensor networks and AI. This beaglebone Board will be useful for me to implement on field based AI to support farmers with decisions in creating a sustainable Agriculture.

Hi,

I'd like to build a in-car safety device, which will monitor driver's attention to the road and create alerts if driver become sleepy.

It will use BB-AI to detect and analyze drivers face, eyes, check if they are open, if they are focused on the road.

The video stream will be captured by a USB and processed by OpenCV.

In addition it can be connected to a gas analyzer to detect if level of CO2 and TVOC are safe.

Thank you for reading my entry!

Sounds like a useful project.

Dubbie

Hi

I would like to create full automatic "e-farmer" me and my family have a lot of animals, from fishes to donkeys - therefore there is a lot of things to do. I would like to connect 5 beaglebone AI via Wifi to my computer to monitor and controle all of the animals (automatic feeding, contolling the light, etc.) I have few little solar panels to power the controllers.

Also because I'm living on a countryside, where is really dark in the night, I have 12-inch Newtonian telescope. Therefore I would like to use one more Beaglenone AI to create "following star" Dobsonian montage.

i Have few step mottors, sollar panels, 3d printer to make housings for the controllers, small cameras, ultrasonic detector.

I hope you can make it possible

Portable ECG machine with CNN based automatic diagnostics

Problem: India, being a vast country, still faces issues in delivering affordable healthcare to all its villages and remote locations. Skilled medical professionals are unwilling to setup medical facilities in these areas which don't have proper infrastructure & connectivity. Timely diagnosis of heart conditions would help the patient seek necessary medical attention from a professional.

Proposed solution: Portable, easy to use, ECG machine which gives diagnosis based on pre-trained Convolutional Neural Network. Using it, ECG data can be gathered by medical assistants to give timely warning to the patient and data can be further analyzed by a medical expert.

Abstract: The Beaglebone-AI possesses the necessary processing power and real-time signal processing abilities to perform both signal analysis and run the CNN. It would make an ideal choice to make a low-cost and portable ECG machine. The display capabilities of the BB-AI is used to interface with a touchscreen based UI. The control abilities of the Cortex-M4s can be used to collect & store ECG data from an multi-channel ADC. The PRUs can efficiently handle buttons and rotary knobs.

How the different components of BB-AI will be used:

-> Dual Cortex-M4: The BB-AI will be paired with a SPI based daughter board. The daughter board will contain Instrumentation amplifiers as well as 12-channel 24-bit ADC. Upto 12 ECG leads can be attached which will be sampled at a rate of 5KSa/s. The 10secs of ECG data will be gathered and stored in memory by the Cortex-M4s using the SPI interface.

-> Dual C66x DSP The signals will be subject to preliminary processing to eliminate noise and enhance important features of the signal such as QRS-complex. The C66x will perform signal segmentation to divide ECG data into beats. The beats will be subjected to Discrete Wavelet Transform (DWT) which will present the data in a Time-vs-Frequency paradigm, which is easier to process by the CNN. It will also collect waveform statistics like amplitude levels and feature intervals.

-> Dual Cortex-A15: They will perform dual roles. They will run the pre-trained CNN for classifying the ECG of the patient. The CNN is trained using data available from free global databases like PhysioNet. Their second role is to run a lightweight touchscreen based UI. The UI will make it easy to view the waveforms from the 12 probes. The UI will expose simple tasks such as maintaining patients record, saving in PDF format, uploading data to cloud etc.

-> Dual PRUs: They will manage real-time control tasks like buttons, rotary knobs and LEDs.

The whole system is battery powered and can upload the ECG data to cloud through WiFi. I estimate the entire system to weigh less than 2Kgs. I also think adding a camera to the system would be beneficial to capture patient image or take pictures of any anomalous body signs. A monochrome USB mini thermal printer can also be added to print ECG reports.

We have already successfully implemented an algorithm to extract features from an ECG signal and classify it. Currently it runs on MATLAB. However, BB-AI looks like the perfect solution to make it embedded. The Raspberry-Pi lacks features to interface with external multichannel ADCs to capture data in real-time. Its nice to see BB-AI possesses the Cortex-M4s to do that.

Hope my idea gets selected!

Sony built a similar feature into their AWS-750 live production console when they launched it a few years ago, which may be of interest.

Anycast Touch Essentials - Episode 9: Auto Camera Tracking & Auto Framing

https://www.youtube.com/watch?v=MpOho93_gpo

It is restricted to cameras using the VISCA RS422 control protocol however and limited in functionality.

I'm not sure whether to try and make some sort of pinhole camera, or lenses based camera, or maybe just stick with putting shapers over the LDR array to create the images.

I am going to try and reduce light leakage by putting each LDR into a hole drilled into some wood which should eliminate any light coming in from the sides. Not sure how deep to make the holes though.

There could well be a problem with different LDRs having different responses to the same amount of light. From the experiment with the 1x8 array, this seems to be minimal. If it is a problem I would be able to calibrate each LDR individually which would be a pain, but it would be achievable. I will be using an ANN to classify images anyway, so small differences in images caused by LDRs having different characteristics should not be too much of a problem.

Dubbie

Is there any particular reason to not use semiconductors?

I think one of the simplest approaches would be to use a pinhole, you definitely have to be very careful with light leaking in, or you won't be able to get much meaningful data. Also its very likely that you will need to calibrate the system, as each photosensor will respond different to light.

In this project I am hoping to help a friend who is blind due to illness. Even Traveling to familiar places is a challenge for her what's more to places she has not been before.  I would like to create a device that is able to recognize objects and inform her about it.

I am hoping to test Beaglebone AI efficiency, accuracy, speed and complexity of initial set. I have tested Object recognition on raspberry pi 4 but it heats up and when It does the accuracy rate goes down.

There are devices out there but the cost are beyond her means. I would like to incorporate a GPS system as that would make it a complete system for her to move around freely.

This sounds familiar... Oh yes I saw this:

Edit media

Dimensions x Small Medium Large Custom

Subject (required) Brief Description Tags (separated by comma) Video visibility in search results Visible Hidden

Parent content

Comment on Project14 | Vision Thing: Beaglebone AI Your Vision Thing Project! Ben Heck’s Auto Tracking Camera Part 1 Episode -- Episode 173

Poster

Upload Preview

Ok

Cancel