Positioning using fixed point cameras - LawnRover


#1

This is an OpenCV and Robotics question – let me know if we need to move it to Robotics.

I’m attempting to build an Autonomous lawn mower to cut a 1.5 acre plot of land.
After researching many ideas on outdoor positioning, I believe that i’ll go with GPS for rough positioning, and fixed-point cameras for more accurate positioning. By using multiple cameras at known positions and angles, along with an easy to detect structure on the rover, I’m hoping to precisely locate the rover (± 20 cm, out to 100m). Graphic shows the concept – I will of course have more cameras. Once position is calculated, it would be sent via RF down to the mower to calculate next steps.

A couple of questions:

  • Has anybody here toyed with a similar concept ?
  • Any experience in detection using OpenCV, or angle calculations based on images?
  • Any suggestions for something which is easily detectable by the cameras? I think could be as simple as a posterboard painted orange. Multiple colored panels on each side could help to determine the heading.
  • I’m also considering just using cameras on the rover, and a neural network approach to guiding it. I think this will be much more difficult up front, but could potentially yield some nice results.

Much appreciated,
Nate


#2

As I recall, they did a crude version of this ( cameras to detect position ) in Circuit Cellar back in the 90s. I think all the back issues are online. Might be of some use, although the tech is vastly better now.

NASA uses high contrast spots ( black on white ) arranged in distinctive patterns to determine orientation and distance on ISS modules and such. A cube on top of the mower with a distict pattern on each side and the top might serve.


#3

#4

Not exactly.

I meant these:

They’re to help align and maneuver the modules - in this case after it was picked up out of the shuttle cargo bay - and get it to it’s docking position. Not exactly fiducials - I know those more from surveying and such. Did not know the cross marks on Apollo era photos were called that as well. ( And these might be as well - have to ping my ISS buddy and ask. )

I think the ones on the ISS modules are part of a computer vision system. They’re doing pattern recognition on the number and relative positions of the spots.


#5

There is a need at the space to geo-locate people then guide them to the room they want.

I have web searched this and no great answers have appeared. There are the $$$$ options but not the reasonable $ options.

The ideal situation is that someone who was trying to find a room would download an APP on their phone and it would guide them to their room.

Variation on a theme would be tablets that lived/charged in the lobby that would do the same thing.

The one that revs my engine is a mouse bot (from Star Wars) or a flying drone that would “escort” you to your destination then return to the lobby awaiting the next request. There are issues with both. Either would need the guest to have a device they would carry so it would know you are “with it”. A guide system that leaves the guest in the lobby is pretty useless.

We had several guests last night who were looking for rooms.

The best technical idea I have come up with are beacons and some classic geometry to geolocate people. Nice on paper, unclear in real space.

FYI - GPS does not have the “fine” positioning for humans or necessarily work that well inside a building. For positioning a missile, +/- a few meters, in the air, is not too big a deal.


#6

Stripes on the wall work remarkably well.


#7

There are several ways to do it but its not cheap. The best I have seen in practice was based off beacons. When used in conjunction with a wayfinding display when button X is pushed and then using a beacon will send guided directions to phones nearby. It can then use secondary beacons to ping when you have arrived at the location you are looking for.

There are systems that work off wifi access points and your phones MAC address, you don’t even have to connect to the access point.

And if you want to get all :tinfoil: there are systems based off facial recognition that can be used to send location to your phone or show on nearby displays when you look at them. These systems are being used in office buildings for employee location, meeting attendance, room usage, and will also monitor mood and participation in meetings.


#8

Idea: LEDs coupled to Arduinos sending short DTMF number sequences. Number corresponds to some location scheme. With a 3 digit code, you get 1000 locations possible. Time per digit is < 60ms for a good detector.

Speaking of detectors, more arduinos with LED receivers and LCD displays. As a user walks around the space, the codes are received and parsed to yield a location. If you want directions, software would have to be developed that implements the rule set. ( I’m here, go that way to get to X. ) That could be done in a lookup table - the number of solutions for each location would be manageable. Add buttons to pick destination from list. Probably could add 4 LEDs for direction as well.

IR LEDs are a good choice for location because they are cheap, as are detectors. And simple beamstops will limit the area over which a particular emitter can be “seen”.

For those that JUST HAVE TO HAVE complicated, make IR to bluetooth devices that will send the codes to the user’s phone. Implement software for Android and IOS. Could be as simple as appending the 3 digit code from bluetooth serial to a fixed URL that has map images tailored for each beacon location.


#9

I do searches on this every 6 months.

There is an APP that might be promising. Microsoft Path Guide.

It needs no hardware beacons. It works by people walking the “path” needed and recording the directions. Then the person who does not know the way chooses the guide to take them to their destination.

This looks promising.

The APP is 11mg which is pretty light weight.

Next time I am at DMS, I will record a few paths and see if I can get someone to follow them with their phone.


#10

You could also manage all this with QR codes on each door and at strategic spots in the hallways. Scan the code, follow link to image of building map with “You Are Here” markers corresponding to the particular code scanned.


#11

Also - apologies to the OP for the thread hijack.


#12

@ADM_Marlow. I am sorry I hijacked this thread. Can you move this series to a new thread please?


#13

I recall a few companies that got embroiled in scandal because their apps were tracking customers’ smartphones using Bluetooth low-energy beacons. Not sure what the granularity was, but I recall they were deployed on streets and in stores.


#14

People didn’t like it, nothing illegal happened and they are still used (a lot) by many companies. Bluetooth and wifi is used to track customer patterns in stores. The data is used to determine store layouts, product placement, and ad placement. The beacons are also used to send stuff to peoples phones but you have to have it enabled and set to receive. It is used for coupons, sales, or links to products when you enter a store or area of a store.

I design and deploy these systems for a living :wink:


#15

Don’t recall the specifics, thus my use of the term “scandal”. There might have been some contentious difference of opinion on terms buried in the boilerplate ULA in the apps in question.

Sure sure. All benefits of using ubiquitous smartphones apps. And also why I generally refuse to install retailer-specific apps - the discounts fail to hit what I consider to be a reasonable value proposition in exchange for such invasive tracking. I’m sure there’s fallback to wifi and cameras to augment these things but that’s different from explicitly giving a retailer permission to painstakingly log my granular location.

As such you could perhaps advise on whether these systems have the potential to work for DMS fine-grained geolocation - presumably with a smartphone app.


#16

Something sort of cool we use these days. Called BrainLab, it locates the tips of a neurosurgical instrument in 3D space. Uses a binocular IR scanner that locates the proximal end of the instrument that has two reflective spots on it. The patient’s CT scan is also registered within the system and the pt’s head has another IR reflector pair in case the surgical bed or scanner is moved.