Vanderbilt NEST ShooterLoc

Slide1: 

Dr. Akos Ledeczi Institute for Software Integrated Systems Vanderbilt University Network Embedded Systems Technology (NEST) Shooter Localization McKenna MOUT Site DBBL, Fort Benning Sep 4, 2003 0900 - 1400

Outline: 

Outline Overview What You Will See Today Technical Approach Performance Caveats/Future Plans

Overview: 

Overview Ad-hoc wireless network of cheap acoustic sensors is used to accurately locate enemy shooters: Nodes detect shockwave and muzzle blast Nodes send back their data to the base station Base station determines shooter location Performance: Average accuracy: 1 meter Latency: 2 seconds Challenges overcome: 7-month research and development time Severely resource constrained cheap nodes Very limited communication bandwidth Bad multipath effects in urban environment CONOPS support: Fast and accurate enemy shooter localization is key in reducing friendly casualties and neutralizing enemy combatants

What You Will See Today: 2D Display: 

What You Will See Today: 2D Display Red circle: Shooter position Red line:  Shot direction Large green circle: Sensor node (good measurement) Small green dot: Sensor Node (no or unused measurement)

What You Will See Today: 3D Display: 

What You Will See Today: 3D Display Red globe: Shooter position Light blue sphere: Sensor node (good measurement) Dark blue sphere: Sensor node (no or unused measurement) Looking south down Main street

Additional Camera Angles: 

Additional Camera Angles Looking south at building B1 Looking east at building B4

Technical Approach: 

Detect TOA of acoustic shockwave and muzzle blast MICA2 mote (UC Berkeley and Crossbow Inc.): Atmel 8MHz microcontroller 4KB data memory Chipcon radio Acoustic sensor board (Vanderbilt): 3 acoustic channels High-speed AD converters FPGA for signal processing: shockwave and muzzle blast detection on board I2C interface to mote 2 AA batteries Timestamp of shockwave and/or muzzle blast sent to mote Motes send data to base station Base station fuses data, estimates shooter position and displays result Middleware services: Time synchronization Message routing Technical Approach

Time Synchronization: 

Time Synchronization Requirements: Sound travels one foot per millisecond Time synch error in the whole network should be less than 1 msec Algorithm: Each mote maintains a separate local and global time Simple integrated leader election Leader broadcasts its time and a sequence number Message is time stamped in the radio stack Receivers update their global time and rebroadcast it If message arrives with known sequence number, it is discarded Motes keep last ten local and global time pairs and perform linear regression If leader is lost, new leader is elected Performance: +/- 60 microseconds error per hop One timesynch round per minute (i.e. one msg per min per mote)

Message Routing: 

Message Routing Requirements: Acoustic event triggers many motes at once All need to get their data to base station with low latency Mote bandwidth: 20-25 messages per second Directed Flood Routing Framework Ad-hoc routing Automatic aggregation Implicit acknowledgments Configurable flooding policy: defines gradient controls retransmission Converge cast to root Performance: When max distance from root is 5, base station receives ~15 measurements in the first second

Slide10: 

t1-vd1 t2-vd2 t3-vd3 t4-vd4 time f(x,y) = 1 / #datapoints in window Sensor fusion outlier

Slide11: 

Sensor fusion Muzzle blast: three dimensional utility function (x,y,z) Shockwave adds three more (azimuth, elevation, bullet speed) Multi-resolution search to find minimum Utilizes all data, does it incrementally as data becomes available Provides estimate, overall error and individual measurement error Discards outliers Multipath effect: Direct line-of-sight motes get real data first Attenuated signals not recognized as shockwave and/or muzzle blast Sensor fusion discards outliers: Simple geometric filtering Utility function

Performance: 

Performance Latency: 2 seconds Accuracy in 2D: 0.64 meter Results below based on 71 SRTA shots from 20 different positions error in meters number of shots Histogram of (x,y) error

Performance: 

Performance error in meters number of shots Histogram of (x,y,z) error Accuracy in 3D: 1.5 meter Results below based on 71 SRTA shots from 20 different positions

Caveats/Future Plans: 

Caveats/Future Plans Caveats #1 Multiple shots: currently 0.4 second separation is required between shots #2 Deployment: current hardware is not weather-proof or shock-proof #3 Silenced weapon or distant shooter: muzzle blast not detected #4 Power usage: currently no power management #5 Scalability: current system scales to ~200 nodes Future Plans #1 Detection should discriminate between different weapons. More intelligent sensor fusion algorithm. #2 New sensor board and packaging needed. #3 Better shockwave detection and sensor fusion #4 Power management is needed. Sentry service is needed. #5 Hierarchical network arrangement is needed