In order to help those with visual impairments navigate streets, college student Satinder Singh has come up with an innovative solution that literally pokes the user in the right direction. 

Singh’s system, called DeepWay, uses a chest-mounted camera to take images of the road that a wearer is walking down, then feeds this information to a laptop for processing. 

If the deep learning algorithm determines that the user needs to move left or right to stay on the path, a serial signal is sent to an Arduino Uno, which in turn commands one of two servos mounted to a pair of glasses to tap the person to indicate which way to walk. Additional environmental feedback is provided through a pair of earphones.

This project is an aid to the blind. Till date there has been no technological advancement in the way the blind navigate. So I have used deep learning particularly convolutional neural networks so that they can navigate through the streets.

My project is an implementation of CNNs, and we all know that they require a large amount of training data. So the first obstruction in my way was a correclty labeled dataset of images. So I went around my college and recorded a lot of videos (of all types of roads and also off-roads). Then I wrote a basic Python script to save images from the video (I saved 1 image out of every 5 frames, because the consecutive frame are almost identical). I collected almost 10,000 such images almost 3,300 for each class (i.e. left right and center).

I made a collection of CNN architectures and trained the model. Then I evaluated the performance of all the models and chose the one with the best accuracy. I got a training accuracy of about 97%. I got roughly same accuracy for all the trained model but I realized that the model in which implemented regularization performed better on the test set.

The next problem was how can I tell the blind people in which direction to move. So I connected my Python program to an Arduino. I connected the servo motors to Arduino and fixed the servo motors to the sides of a spectacle. Using serial communication I can tell the Arduino which servo motor to move which would then press to one side of the blind person’s head and would indicate him in which direction to move.

A demo of DeepWay can be seen in the video below, while code for this open source project is available on GitHub.

The “Navigation Thing“ was designed and built by [Jan Mrázek] as part of a night game activity for high school students during week-long seminar. A night-time path through a forest had stations with simple tasks, and the Navigation Thing used GPS, digital compass, a beeper, and a ring of RGB LEDs to provide a bit of “Wow factor” while guiding a group of students from one station to the next. The devices had a clear design direction:

“I wanted to build a device which a participant would find, insert batteries, and follow the beeping to find the next stop. Imagine the strong feeling of straying in the middle of the night in an unknown terrain far away from civilization trusting only a beeping thing you found. That was the feeling I wanted to achieve.”

The Navigation Things (there are six in total) guide users to fixed waypoints with GPS, a digital compass, and a ring of WS2812 LEDs — but the primary means of feedback to the user is a beeping that gets faster as you approach the destination. [Jan] had only four days to make all six units, which was doable. But as most of us know, delivering on a tight deadline is often less about doing the work you know about, and more about effectively handling the unexpected obstacles that inevitably pop up in the process.

The first real problem to solve was the beeping itself. “Beep faster as you get closer to the destination” seems like a simple task, but due to the way humans perceive things it’s more complex than it sounds. We perceive large changes easier than small incremental ones, so a straight linear change in beep frequency based on distance doesn’t work very well. Similar problems (and their solutions) exist whether you’re controlling volume, brightness, or just about anything else that humans perceive. Instead of encoding distance as a beep frequency, it’s much more effective to simply use beeps to signal overall changes: beep noticeably slower as you move away, but beep much faster as you get close.

The other interesting problems were less straightforward and were related to the digital compass, or magnetometer. The first problem was that the piezo buzzers [Jan] sourced contained no actual piezo elements. They contained magnets – which interfered with the operation of the digital compass. After solving that, still more compass problems arose. When testing the final units in the field, the compass readings were not as expected and [Jan] had no idea why.

After careful troubleshooting, the culprit was found: the AA cells on the other side of the circuit board. Every AA cell has a faint (and slightly different) magnetic field, and the proximity and placement of the cells with respect to the magnetometer was causing the deviation. Happily, the fix was simple once the problem was understood: calibrate the compass every time new batteries are inserted.

If you’re interested in the Navigation Thing, check out the github repository. And on the topic of actual piezoelectric devices, piezos are implemented in a variety of clever ways. There are even piezo transformers and piezo vacuum pumps.

There are plenty of GPS navigation units on the market today, but it’s always fun to build something yourself. That’s what [middelbeek] did with his $25 GPS device. He managed to find a few good deals on electronics components online, including and Arduino Uno, a GPS module, and a TFT display.

In order to get the map images on the device, [middelbeek] has to go through a manual process. First he has to download a GEOTIFF of the area he wants mapped. A GEOTIFF is a metadata standard that allows georeferencing information to be embedded into a TIFF image file.  [middelbeek] then has to convert the GEOTIFF into an 8-bit BMP image file. The BMP images get stored on an SD card along with a .dat file that describes the boundaries of each BMP. The .dat file was also manually created.

The Arduino loads this data and displays the correct map onto the 320×240 TFT display. [middelbeek] explains on his github page that he is currently unable to display data from two map files at once, which can lead to problems when the position moves to the edge of the map. We suspect that with some more work and tuning this system could be improved and made easier to use, of course for under $25 you can’t expect too much.

Bike sharing systems like New York's Citi Bike may be taking off, but it's doubtful that many participants can find every station without checking a map. Thankfully, Adafruit has unveiled a smart helmet project that could help at least a few of those riders get to their destinations while keeping their eyes on the road. The DIY effort feeds locations to an Arduino-based Flora board and its positioning add-ons, which in turn use a string of NeoPixel LEDs on the helmet as turn indicators. Commuters just have to watch for blinking lights to know where to go next. While the system isn't easy to set up when cyclists have to manually enter coordinates, it is flexible: the open-source code lets it adapt to most any bike sharing system or headpiece. As long as you can get over looking like a Christmas tree on wheels while you navigate, you can build a smart helmet of your own using the instructions at the source link.

Filed under: GPS, Transportation, Alt

Comments

Source: Adafruit