38
Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2 Shen, Brostow, Cipolla University of Cambridge

Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Embed Size (px)

Citation preview

Page 1: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Toward Automatic Blood Spatter Analysis in Crime

Scenes

Gabriel Brostow, 13 June, 2006

Shen, Brostow, CipollaUniversity of Cambridge

Page 2: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Bloodstain Categories*

• Passive Bloodstains

• Projected Bloodstains– Low / medium / high velocity impact:

caused by force applied to a blood source

• Transfer/Contact Bloodstains

*International Association of Bloodstain Pattern Analysts

Page 3: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge
Page 4: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Image by Kevin Maloney

Page 5: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Point of Origin Localization

Page 6: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

String Method

Page 7: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Support Software

From BackTrack software by A. L. Carter, 2001 version

Page 8: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Automation Goals

1. Estimate impact angles for 1 spot

2. Estimate 2D origin of impact

3. Estimate 3D origin of impact

Page 9: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Automation Goals

1. Estimate impact angles for 1 spot

2. Estimate 2D origin of impact

3. Estimate 3D origin of impact

Page 10: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge
Page 11: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Experimental Setup

Page 12: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Results: Impact Angles

Page 13: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Primary vs. Secondary Stains

Bevel & Gardner, 2001

Page 14: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Filter for Outliers

Page 15: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Automation Goals

1. Estimate impact angles for 1 spot

2. Estimate 2D origin of impact

3. Estimate 3D origin of impact

Page 16: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

2D Multi-Spot Analysis

• Experiment– Blunt force impact– True origin:

• diameter 6cm• height 22cm

Page 17: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Strings in the form of vectors

x

y

Page 18: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Intersections

Page 19: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Convolve with Gaussian

Page 20: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Threshold on Distance

Page 21: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Image rectification

• Generate synthetic view

• Homography:

Page 22: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Image rectification

Page 23: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Automation Goals

1. Estimate impact angles for 1 spot

2. Estimate 2D origin of impact

3. Estimate 3D origin of impact

Page 24: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge
Page 25: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge
Page 26: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge
Page 27: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge
Page 28: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge
Page 29: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Height Estimation

• Triangulation– H = tan() * distance– True height 22cm,

estimated height 19cm

• Advanced– Unknowns– Speed, distance, air

resistance and gravity

Page 30: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Height Estimation

• Triangulation– H = tan() * distance– True height 22cm,

estimated height 19cm

• Model in future– Speed, distance, air

resistance and gravity

Page 31: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Findings

• Demonstrated accuracy of 1-spot analysis

• 2D Origin of Impact Estimation

• Overhead crime-scene visualization

• Groundwork for 3D string method automation

Page 32: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Future Work

• Real blood images

• 3D projectile trajectory modeling

Page 33: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge
Page 35: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge
Page 36: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Related work

• Interpretation of Bloodstain Evidence at Crime Scenes – Eckert & James, 1998

• Bloodstain Pattern Analysis– Bevel & Gardner, 2001

• Blood Dynamics– Wonder, 2001

• The Directional Analysis of Bloodstain Patterns, Theory and Experimental Validation– A.L.Carter, 2001

Page 37: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Alternative bloodstain ellipse fitting

• Alternative ellipse fitting algorithm– Ellipse growth– Erosion, median filter

and dilation

Page 38: Toward Automatic Blood Spatter Analysis in Crime Scenes Gabriel Brostow, 13 June, 2006 Shen, Brostow, Cipolla University of Cambridge

Manual vs. Automatic

• Current pipeline:– On-site measurements– Physical strings

construction– Qualitative estimation

of origin

• Automatic pipeline:– Image processing– Strings in the form of

equations stored on computer

– Quantitative estimation of origin using error functions