an optimised frame work for maximum activity … · an optimised frame work for moving target...

30
AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR # Aeronautical Development Establishment(ADE), Defence Research and development Organization(DRDO), Bangalore - 560075 # Siddaganga Institute of Technology(SIT), Tumkur

Upload: others

Post on 30-Jul-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR

UAV APPLICATION

Md. Shahid, Pooja HR#

Aeronautical Development Establishment(ADE), Defence Research and development Organization(DRDO),

Bangalore - 560075

# Siddaganga Institute of Technology(SIT), Tumkur

Page 2: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Agenda

ADE: UAV Scenario

Introduction

Problem Statement

Acquisition and Tracking

Proposed Frame work

Evaluation Criteria

Results Analysis

Conclusion

Scope of future work

Discussion

Page 3: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

ADE: UAV Scenario

LAKSHYA – Aerial target NISHANT – TUAV RUSTOM-1 UAV

Page 4: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Introduction

Electro Optics (EO) Payload

DTV Camera

FLIR Camera

Electronics package

Unmanned Aerial Vehicles (UAVs) are increasingly being used for reconnaissance and surveillance.

GROUND CONTROL STATION (GCS)

Page 5: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Problem Statement

Acquisition: – Moving target detection under platform disturbances and delays.

Tracking: – Tracking target/vehicle independent of speed/maneuvering.

– Not limited to number of targets.

Target Acquisition

Target Tracking

Manual Automatic

Page 6: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Why moving target detection?

Acquiring moving targets from airborne platform is difficult task due to associated delays.

– Video Downlink: 200 mSec

– Commands uplink: 400 mSec

– Object displacement: ~ 100 pixels (in acquisition)

Ground Control Station(GCS) Antenna Vehicle

Video downlink (200 mSec)

Command uplink

(400 mSec)

Page 7: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Acquisition Problem

Page 8: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Tracking

Page 9: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Proposed Frame work for Acquisition

Input

Video

Moving targets

detected video

* MATLAB 2013a and its tool boxes

Interest point Detection

(Eligibility Criteria)

Registration (Scene lock)

Background Subtraction

Clutter reduction

Page 10: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Interest Point Detection

Definition: – Local image structure around the interest point is rich in terms of

local information contents.

Examples : Corner, blob, ridge, edge etc.

Corner Detection Techniques:

– Harris Detector.

– Moravec Operator.

– Features From Accelerated Segment Test (FAST).

– Median Method.

Interest point Detection

(Eligibility Criteria)

Registration (Scene lock)

Background Subtraction

Clutter reduction

Page 11: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Corner detection results

(a) (b) (a) (b)

(c) (d)

Figure : Corners detected by (a) Harris Detector (b) Moravec Operator (c) FAST algorithm (d) Median algorithm

Page 12: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Complexity Analysis

Complexity in terms of computations and memory (per pixel), is as follows

ALGORITHM

Multiplication operation

Arithmetic operation

Division/ Comparision

operation

Memory requirement

Robustness

Harris Detector

49 43 - 18

Excellent

Moravec Operator

72 136 - 1

Moderate

Fast Algorithm

1 18 - 2

Good

Median Algorithm

- 6 1/26 1 Moderate

Page 13: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Eligibility Criteria

A3(2x2)

A2(2x2)

A1(2x2)

A0(2x2)

Restricts the eligible candidates(pixels) to be under process for further corner detection.

Considering the pixel as a centre for its 5x5 size block A and its four sub-blocks.

Sub-blocks are of 2x2 size each

Threshold : D2 > Mean(A)

Reduces computational burden significantly.

3210 2 AAAAH

0213 2 AAAAV

222 VHD

Page 14: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Robustness of eligibility criteria

Eligibility criteria is robust under

– Translation

– Rotation

– Scaling

– Noisy environment

Computational saving

Algorithm/ Technique

Processing time (all pixels)

Processing time (eligible pixels)

Computational Saving

Harris Detector

~1.5458 sec ~0. 5825 sec 62%

Moravec Operator ~0.5449 sec

~0.2075 sec 61%

FAST Algorithm

~0.1059 sec

~0.0399 sec

62%

Page 15: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Results of eligibility criteria

(a) (b) (c) (d)

(e) (f) (g) (h)

Figure : (a) Original image of size 256x256. (b) Corners detected by Harris Detector. (c) Rotated image (-100).

(d) Corners detected for rotated image. (e) Image affected by Gaussian noise. (f) Corners detected for Image affected by Gaussian noise. (g) Original image resized to 128x128. (h) Corners detected for resized image.

Page 16: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Results of eligibility criteria(UAV image)

(a) (b) (c) (d)

(e) (f) (g) (h) Figure : (a) Original image of size 256x256. (b) Corners detected by Harris Detector. (c) Rotated image (-100). (d)

Corners detected for rotated image. (e) Image affected by Gaussian noise. (f) Corners detected for Image affected by

Gaussian noise. (g) Original image resized to 130x130. (h) Corners detected for resized image.

Page 17: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Registration

Why Registration?

– Arresting the background against platform movement

– Required for moving or dynamic platform

Assumptions

– Background forms most part of the scene

– Background interest points moves slower

than foreground

Multiple target tracking

– Circularization and correlation matching

– Restricting to least movement, r2 = x2 + y2

– Proportional weightage X = (4*x1 + 3*x2)/7,

Y = (4*y1 + 3*y2)/7

– Discarding poor target & Best target weigh more

Unregistered

video clip

Registered

video clip

Interest point Detection

(Eligibility Criteria)

Registration (Scene lock)

Background Subtraction

Clutter reduction

Page 18: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Background Subtraction

Background subtraction

– Key aspect of the frame work

Type of backgrounds

– Dynamic backgrounds

– Gradual illumination changes

– Sudden illumination changes

– Moved object

– Shadows

Various methods

– Pixel or region based methods

– Parametric or nonparametric methods

– Recursive or non-recursive methods

Interest point Detection

(Eligibility Criteria)

Registration (Scene lock)

Background Subtraction

Clutter reduction

Page 19: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Background Subtraction Methods

Bayesian histogram

Morphological filtering

Sigma-delta(∑-∆) motion detection

Visual Background Extractor (ViBe)

Page 20: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Static platform

(a) (b) (c)

(d) (e) (f)

Figure : a) Input video frame of static camera. b) Ground truth. c) Bayesian histogram. d) Morphological Filtering. e) ∑-∆ motion detection. f) ViBe method. Courtesy: video sequence Highway II (available at http://cvrr.ucsd.edu/aton/shadow/data/highwayII-raw.avi)

Page 21: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Moving platform

(a) (b) (c)

(d) (e) (f)

Figure : (a) Input video frame of UAV. (b) Ground truth. (c) Bayesian histogram. (d) Morphological Filtering. (e)∑-∆ motion detection. f) ViBe method. (Note: Clutter has not been removed)

Clutter

Page 22: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Clutter reduction algorithm

Background subtracted image

Remove the area containing less than 200 pixels

Find boundaries

Find, A o = Area occupied by object

& A B = Filled area of bounding box

Fill Ratio, A = A o/ A B

Aspect Ratio

A r =Width/Height

( A > 0.5) ?

Clutter

No

Moving Objects

(3.5 > A r > 1 )? No

Yes

Yes

Background Subtraction 1

2

Interest point Detection

(Eligibility Criteria)

Registration (Scene lock)

Background Subtraction

Clutter reduction

Page 23: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Evaluation Criteria

Subjective: Visual inspection

Objective :

– Percentage of Correct Classification(PCC)

– False Positive Rate (FPRate)

Where,

True positives(TP): Number of correctly detected foreground pixels

False positives(FP): Number of background pixels incorrectly classified as foreground

True negatives(TN): Number of correctly classified background pixels

False negatives(FN): Number of foreground pixels incorrectly classified as background

FNFPTNTP

TNTPPCC

TNFP

FPFPRate

Page 24: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Results Analysis

Page 25: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Conclusion

MATLAB helped us all the way to develop this frame work for real time UAV application

Extensively utilized following MATLAB Tool boxes

– Computer vision system tool box

– Image Processing tool box

– DSP System Processing tool box

– Statistics tool box

Quick study of various methodologies

Not limited to number of moving targets

Complexity is independent of target speed

Reduced time to develop this framework

Page 26: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Scope of future work

Immediate:

– Target merged to clutter

– Fill ratio criteria

(a) (b)

Figure : (a) Targets merged with clutter. (b) Target failing to fill ratio criteria.

Next:

– Clutter is more due to rolling

– Replacement of Morphological operation

Page 27: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

References

• O. Barnichand M. Van Droogen broeck, "ViBe: A Universal Background Subtraction Algorithm for Video Sequences", IEEE Transactions on image processing, Vol 20, no.6, June 2011.

• A Robust and Computationally Efficient Motion Detection Algorithm Based on ∑-∆Background Estimation. A. Manzanera J. C. Richefeu. ENSTA/LEI, 32 Bd VictorF-75739 PARIS CEDEX 15, july 6, 2011.

• “FASTER and better: A machine learning approach to corner detection” in IEEE Trans. Pattern Analysis and Machine Intelligence, Edward Rosten, Reid Porter and Tom Drummond, vol 32, pp. 105-119, 2010.

• Saad Ali and Mubarak Shah, COCOA - TRACKING IN AERIAL IMAGERY. Computer Vision Lab,School of Computer Science,University of Central Florida, 2006.

• A Background Subtraction Model adapted to Illumination changes. Julio Cezar Silveira Jacques Jr., Claudio Rosito Jung and Soraia Raupp Musse, IEEE Transactions, 1-4244-0481-9/06/$20.00 C 2006.

• C. Harris and M. Stephens. A Combined Corner and Edge Detector. Proc. Alvey Vision Conf., Univ. Manchester, pp. 147-151, 1988.

• H. P. Moravec. Towards Automatic Visual Obstacle Avoidance. Proc. 5th International Joint Conference on Artificial Intelligence, pp. 584, 1977.

Page 28: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Acknowledgement

Aeronautical Development Establishment (ADE),

Defence Research and Development Organization (DRDO),

Bangalore, India.

Department of Electronics & Communication Engg.,

Siddaganga Institute of Technology (SIT), Tumkur, India.

CoreEL Technologies, Bangalore, India.

Page 29: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

Discussion

Page 30: AN OPTIMISED FRAME WORK FOR MAXIMUM ACTIVITY … · AN OPTIMISED FRAME WORK FOR MOVING TARGET DETECTION FOR UAV APPLICATION Md. Shahid, Pooja HR# Aeronautical Development Establishment(ADE),

FOR MORE DETAILS CONTACT

Md. Shahid,

[email protected]

Pooja H.R,

[email protected]