Documentation : Hardware | Software | Tutorials    Wardrobe : Jackets | Applications
Papers : Notes | Proposals | Research    Administration : Stock | Minutes | Progress

CITY Project

Workshop: 7th-15th January 2003

Notes, images and data collected by Cliff

n.b. other work was also carried out by other attendees at the workshop - primarily Ian and Chris, but also Matthew, Martin, Marek and David. Thanks also to Anthony (and UCL colleagues for their help).

On this page :-

Reference Co-ordinates
Ultrasonic Rx Comparisons
Tracking in the Streets
Pinger Performance
Buchy2Mack
WiFi Antenna

 

ian_hack.jpg (26318 bytes)


Summary

During this workshop the objective of tracking a user from Buchanan Street to the Lighthouse Mackintosh Room was achieved. The overall accuracy was not measured, however an indicative figure of 15m 95% of the time is not unreasonable (excepting, of course, in the Mack Room). Testing was also carried out using an upgraded ultrasonic receiver in the Mack Room with favourable results. 

In addition considerable work was carried out on post workshop analysis of the position data, including the conversion between WGS84 and OSGB.

The following actions are recommended for the next workshop or visit :-

  1. modify ultrasonic analysis software to accommodate reflections from rhs wall, and to extract 'y' positions from sensors on top of screen;
  2. revert to limited range ping in Mack Room;
  3. incorporate real time raster checking into pMack;
  4. experiment further with sensor fusion algorithms;
  5. introduce location pinger validation algorithm;
  6. identify more appropriate pinger locations in Lighthouse (away from external walls).

Reference Co-ordinates

Mack Room cartesian 'origin' calculated to be at 55deg 51.5853min N., 4deg 15.3318min W. from dGPS data and superimposed images collected at previous workshop (see below).

Manual readings from Explorer Map 342 and UCL bitmap:

OS (Airy1830)

OS

OS (UCL bitmap)

Lat

55deg 51.5676N

65250

65250

Lon

4deg 15.4286 W

58925

58928

dGPS Readings converted to OS using Jeeps, and resulting (negligible) correction offsets:

dGPS (WGS84) Jeeps OS Offset (to UCL)
Lat 55deg 51.5853 N 65248 -2m
Lon 4deg 15.3318 W 58929 +1m

OS converted to WGS84 using Jeeps, and resulting correction offsets:

OS (UCL bitmap) Jeeps WGS84 Offset (to dGPS)
Lat 65250 55deg 51.5304 N -0.0549min
Lon 58928 4deg 15.1007 W -0.2311min

So, Jeeps GPS(WGS84) to OS appears accurate (within ~2m), but Jeeps OS to WGS84 is out by ~100s of metres (later - Ian finds a bug!).

n.b. Why is there a difference between OS lat/lon and WGS84 lat/lon? The Earth is not a sphere - to cope with it's irregular shape, national maps use a coordinate system that best fits Britain only (Airy 1830). GPS uses the "World Geodetic System 1984" (WGS84). When we use our dGPS readings we are not taking into account the Airy 1830 "distortion" used for OS maps. 

Lots more conversions here, and the GlasgowRasters page is handy too!

super.gif (304351 bytes)

 

 

Calculation of lat/lon (WGS84) of origin in Mack Room

 

Pos3

Pos2

Origin

x (N/S)

97

249

359

y (E/W)

204

468

604

lat

51.5881

51.5865

51.5853

lon

15.3161

15.3265

15.3318

 


Ultrasonic Rx Comparisons

A monostable has been added to the ultrasonic processing circuitry, and the decoding software has also been tweaked to filter spurious signals - results below using the standard Mack Room test trail (nb with zoning applied).

old_rx.gif (10248 bytes)

Old Rx - less sensitive

new_rx.gif (11391 bytes)

New Rx - more accurate with interesting anomalies

  The new receiver clearly is obtaining more consistent readings, however it is also picking up second reflections from the rhs wall, and there is also a problem with missing 'y' data in the centre of the room (causing the straight lines with constant y). Both these effects should be capable of being resolved with some minor tweaks to the software - at the next workshop!


Tracking in the Streets

The following plots were obtained using GPS Receiver (Garmin GPS25 + ST Ping Antenna) and deduced (or dead) reckoning (DR) data derived from a pedometer and a handheld compass. The data was sent wirelessly to the pinger receiver as used for the ultrasound installation in the Mack Room. The test path started at the intersection of Buchanan Street and Gordon Street (outside TGI Fridays); went down Buchanan Street; turned right into Argyle Street, and right again up Union Street; then back to the start.

The first column of images shows the raw data superimposed approximately on a Glasgow streetmap - GPS readings are dark blue and DR readings are pink; the second column shows the effect of dynamically adusting the DR readings (i.e. stride and compensating for lost wireless data). The third column shows a first attempt at automatically reconciling the readings with Ordnance Survey building data (thanks Ian and Anthony) - the map is based on OS Data, i.e. is more accurate than the first two columns, the white pixels are GPS readings and the grey either DR or adjusted GPS. The algorithm used to move readings out of buildings takes into account the last known position of the sensor.

Raw Data

DR Updated (increased stride)

Raster Corrected (nearest external)

1057_rawmap.gif (29700 bytes)

1057_mapfix.gif (29970 bytes)

1057_rc.gif (2496 bytes)

1106_rawmap.gif (29440 bytes)

1106_mapfix.gif (30426 bytes)

1106_rc.gif (2521 bytes)

1115_rawmap.gif (29558 bytes)

1115_mapfix.gif (30251 bytes)

1115_rc.gif (2464 bytes)

More raster checking:

Key: Cyan - unadjusted GPS; Green - adjusted GPS; Magenta - DR; Red - adjusted DR.

Readings inside buildings moved to nearest external point Readings inside buildings moved to external point nearest to previous reading.

1057_n.gif (2864 bytes)

1057_p.gif (2862 bytes)

1106_n.gif (2888 bytes)

1106_p.gif (2868 bytes)

1115_n.gif (2850 bytes)

1115_p.gif (2844 bytes)

The above images were generated using DR data already processed in the Jornada. The following images, generated using raw data, enable the DR to be based on raster corrected GPS.

Key: Green - unadjusted GPS; Light green - adjusted GPS; Orange - DR; Red - adjusted DR.

Raw GPS

Raster Corrected GPS

raw_gps.gif (2748 bytes)

raster_corrected.gif (2744 bytes)

Raster Corrected GPS with DR

Raster corrected GPS with raster corrected DR.

corrected_with_DR.gif (2828 bytes)

DR_corrected.gif (2828 bytes)

Encouraging results with speed estimation of the DR being the main source of error. However the two images below show much poorer performance:

alt_1.gif (2870 bytes)

alt_2.gif (2808 bytes)

Unfortunately the data collected for the above images may have been affected by erratic user behaviour (caused by a downpour!). Nevertheless it demonstrates the fragility of the techniques used. A GPS error of 10m is sufficient to place the receiver in the next street. A handheld compass is also subject to the user's extraneous movements. We also see a problem with the user getting stuck in dead ends (off Buchanan St.). 

 


Pinger Performance

As part of the Buchy2Mack tests, we installed Location Pingers (beacons) on each floor of the Lighthouse. On the first floor the pinger was just inside the main doors on the reception desk/podium. On the first floor the pinger was placed in the leaflet dispenser at the top of the escalator. On the second floor it was placed in the window alcove, and on the third floor on top of the external door opposite the entrance to the Mack Room. The pingers were set to an approximate range of 10m.

The table below shows three plots of three tests (different colours) with different filtering applied. The vertical axis corresponds to the floors of the Lighthouse and the horizontal is a count of the pings received.

Raw pings - showing considerable jumping between floors. ping_raw.gif (12267 bytes)
Filtered pings - requires two consecutive pings for valid location fix.

Less between floor jumping - 'pink' plot is good representation, though location outside Mack Room (line 4) is not identified - this may be good reason to revert to a limited range pinger inside the Mack Room. 

ping_filt2.gif (9356 bytes)
Filtered pings - requires three consecutive pings for valid location fix.

Delay probably excessive, and some floors not recognised.

ping_filt3.gif (7130 bytes)

Observations:

1) Filtering the pings by requiring a second validation ping considerably reduces the spurious jumping between ping locations caused by overlapping ping boundaries. The penalty is (usually) a one second delay in reporting a location fix.

2) Pinger placement is critical - the ground floor pinger clearly has a range which extends to the third floor. If more suitable locations cannot be found the pingers will require modification.


Buchy2Mack

This section shows the results of tracking a user from outside in Buchanan St (GPS), moving into Mitchell Lane (DR), entering the Lighthouse and using the escalators to climb to the third floor Mack Room (Location Pingers); and then walking round the Mack Room (Ultrasonics).  

Outside

Coloured pixels (see key above) show path down Buchanan St (GPS), then turning into Mitchell Lane with DR readings. There also appear to be GPS fixes being obtained while in Mitchell Lane - undoubtedly of poor quality.  1216_B2M.gif (2413 bytes) snap0012.jpg (65290 bytes)
Recorded path from Buchanan St to the Lighthouse Mitchell Lane

Inside the lighthouse

The figure below shows the pinger readings taken inside the Lighthouse - the three colours represent the three filtering options applied to the same set of data (blue - raw; pink - single additional validating ping; yellow - double additional validating pings). It should be noted that spurious pings were received while we were in Mitchell Lane - probably due to the proximity of the pingers to the external walls.

ping_comb.gif (10820 bytes)

Around the Mack Room

... and a quick once round the Mack Room

1216Mack.gif (7558 bytes)


WiFi Antenna

WiFi antenna placement outside Mack Room window for most excellent coverage of Mitchell Lane - from NCP in Mitchell Street to two shop widths on other side of Buchanan Street (and 1km into the sky):-

 

n.b without the additional antenna WiFi coverage did not extend beyond the Buchy St end of Mitchell Lane.

snap0006.jpg (50333 bytes)


Cliff