20161209 JAWS-UG AI支部 #2 LT : Moving story of AWS/ML beginner engineer
1 like561 views
The document discusses using machine learning to identify indoor location based on Wi-Fi access point (AP) signal strength. It describes an engineer's initial trial using training data consisting of Wi-Fi AP info and signal strengths along with floor indexes to predict location. The engineer refined the training data and data structure, improving results from the initial to third iteration. The engineer encourages others to experiment with indoor location identification using readily available data from Wi-Fi APs and other sensors through low-cost machine learning methods.
20161209 JAWS-UG AI支部 #2 LT : Moving story of AWS/ML beginner engineer
- 1. Moving story of AWS/ML beginner engineer Fujitsu Software Technologies Global Knowledge Center 9 Dec. 2016 根木 neki.atsushi@jp.fujitsu.com
- 2. • Development / Porting • Embedded Linux Kernel • Wi-Fi MW/Drv for Android • SSD firmware
- 3. Q. Is it possible to identify indoor location by using Wi-Fi AP signal strength ?
- 4. Yes, it is. • Cisco Meraki • CSI based indoor location • CSI : Channel State Info and so on..
- 5. But… • Difficult !! ( For me !! ) • You’ll need special hardware, skill and privilege.
- 6. Cutting corners ! • Using Machine Learning • To calculate position from AP signals • I’ve just started this trial since this week.
- 9. Training!
- 10. Training Data 1 mac_ 001 ch_0 01 rssi_ 001 ssid_ 001 mac_ 002 ch_0 02 rssi_ 002 ssid_ 002 mac_ 003 ch_0 03 rssi_ 003 ssid_ 003 mac_ 004 ch_0 04 rssi_ 004 ssid_ 004 mac_ 005 ch_0 05 rssi_ 005 ssid_ 005 mac_ 006 ch_0 06 rssi_ 006 ssid_ 006 mac_ 007 ch_0 07 rssi_ 007 ssid_ 007 mac_ 008 ch_0 08 rssi_ 008 ssid_ 008 mac_ 009 ch_0 09 rssi_ 009 ssid_ 009 mac_ 010 ch_0 10 rssi_ 010 ssid_ 010 floor 84:8 E:DF: ED:1 B:B1 11 -19 nekia p 00:23 :26:7 3:58: 29 6 -58 FST - WIFI - NET WOR K 00:23 :26:7 3:58: 2E 6 -59 00:23 :26:7 3:58: 28 6 -60 00:23 :26:7 3:58: 2F 6 -61 58:7 F:66: 0A:4 6:2E 3 -65 W01_ 587F 660A 462E E0:1 8:77: 70:6 E:50 9 -65 fstso umu 00:23 :26:7 7:F9: 29 1 -67 FST - WIFI - NET WOR K 00:23 :26:7 7:F9: 2F 1 -68 00:23 :26:7 7:F9: 2E 1 -68 601 84:8 E:DF: ED:1 B:B1 11 -19 nekia p 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 601 84:8 E:DF: ED:1 B:B1 11 -19 nekia p 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 602 84:8 E:DF: ED:1 B:B1 11 -19 nekia p 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 602 84:8 E:DF: ED:1 B:B1 11 -19 nekia p 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 603 84:8 E:DF: ED:1 B:B1 11 -19 nekia p 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 603 84:8 E:DF: ED:1 B:B1 11 -19 nekia p 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 603 84:8 E:DF: ED:1 B:B1 11 -25 nekia p 00:23 :26:7 3:58: 2F 6 -55 00:23 :26:7 3:58: 2E 6 -57 00:23 :26:7 3:58: 29 6 -58 FST - WIFI - NET WOR K E0:1 8:77: 70:6 E:50 9 -63 fstso umu 00:23 :26:7 3:58: 28 6 -63 58:7 F:66: 2A:1 4:7B 11 -63 W01_ 587F 662A 147B 00:23 :26:7 7:F9: 29 1 -66 FST - WIFI - NET WOR K 58:7 F:66: 0A:4 6:2E 3 -66 W01_ 587F 660A 462E DC:F B:02: DD:A 7:E0 10 -72 WTH DIYK ETB KM- 24G 604 • Just serializing the following: • 10 Wi-Fi AP Info • floor index <-- this is target !
- 11. Result 1
- 12. Training Data 2 • Gathered more training data. • about 10 times more
- 13. Training Data 2 • Gathered more training data. Training 1
- 14. Result 2
- 15. Training Data 3 • Changed data structure. 84:8E:DF:ED:1B:B1 FA:8F:CA:32:CF:88 00:23:26:73:58:29 00:23:26:77:F9:29 00:23:26:77:F9:28 00:23:26:77:F9:2F 00:23:26:73:58:2E 00:23:26:73:58:28 00:23:26:73:58:2F floor -19 -44 -51 -66 -63 -65 -61 -59 -60 601 -19 -46 -49 -61 -255 ・・・ -255 -61 -59 -60 601 -19 -43 -50 -66 -255 -255 -51 -50 -60 601 -18 -47 -47 -62 -255 -255 -60 -60 -59 601 1 data consist from about 130 AP RSSI + floor
- 17. Further research • Launch an application to track location. • Add other sensor infos as training data • BLE beacon • Geomagnetic field sensor
- 18. It’s extremely EASY!! Let’s dive in ! • You can just start it without any special knowhow. • You don’t need much money (it takes only $5, so far). • You don’t need any special machine.
- 21. Result 1
- 22. Result 2
- 23. Result 3