Project CC: Bullets and Cartridge Cases

# Project CC: Bullets and Cartridge Cases
### 2019-06-10

---

Graduate Students

- Ganesh Krishnan
- Kiegan Rice
- Charlotte Roigers
- Joe Zemmels

Undergraduates

- Talen Fisher (fix3p)
- Mya Fisher, Connor Hergenreter, Carley McConnell, Amy  (scanner)
- REU-sprite: Syema, Tiger, Emmanuelle
- REU-csafe: Andrew, Molly

---
class: primary
# Awards and other News

- Kiegan received a first prize for her poster at All Hands

- Talks at AFTE (Kiegan and Heike) went well, lots of good questions
- More bullets from Prof Hamby: Clones of HS 224 (complete and labelled)

---
class: primary
# Bullet projects - Big picture

- **data collection**

- **computational tools**

- matching lands:

1. crosscut identification
    2. **groove location**
    3. curvature removal
    4. alignment of signatures
    5. feature extraction
    6. matching with **trained Random Forest**

- **analysis of results**

- **communication of results**

---
class: primary
# Update from the data collection

- **scans from bullet lands (about 20,000 total)**

- **LAPD: 4 bullets per barrel for 526 out of 626 firearms**
    - **Variability study: ~2000 scans**
    - Hamby Sets 10, 36, 44, 224, and a clone (35 bullets each)
    - Houston test sets (6 kits with 25 bullets each)
    - Houston persistence: 8 barrels with 40 fired bullets each
    - St Louis persistence: 2 barrels with 192 fired bullets each
    - most of the CSAFE persistence study 
    
- **and cartridge cases**

- DFSC (about 2000)
    - getting ready to scan cartridges for CSAFE persistence
    
- **shooting range**
    
    - **we were out on the range last week: barrels #4 to #7 (in record time!)**
    - **we are planning to go out to the range one more time to finish up the persistence study**

---
class: primary
# Shooting range

---
class: primary
# Groove identification

Two papers in preparation:

- (Kiegan) Journal of Forensic Sciences paper on initial groove ID methods  
    - Getting into right format, submitting today or tomorrow!  
    
- (Kiegan and Nate) FSI Paper on advanced groove ID methods
    - Two-Class Classification Method
    - Bayesian Changepoint Method (Nate)
    - Pairwise results comparison on 3 bullet test sets   
    - Cleaning up, getting ready for submission  
    
- Future Work: Hough Transformation (Charlotte) 
---
class: primary
# Computational tools

- **fix3p**:  Chrome extension by Talen Fisher (updated)

- **x3ptools**

- **grooveFinder**

- implementation of various techniques to identify groove locations
    - github: [heike/grooveFinder](https://heike.github.io/grooveFinder/)
    
    - pull request to `imager` to remove plyr dependency (Susan) - merged yesterday!

---
class: inverse
# Charlotte

---
class: primary
# Hough Transforms

- A computer-vision algorithm that detects aligned points in an image
- Two points appear on the same line in an image will have intersecting lines in the feature space.
- Strong edges will have a lot of points along it, therefore a lot of intersections at that specific point.
  
.center[
<img src = "charlotte/06_10_2019_Update/hough_parametrization.png" width = 50% height = 50% >
]

---
class: primary
# Hough Transform
.center[
<img src = "charlotte/06_10_2019_Update/Hamby_252_Bullet1_Land3_Hough.png" width = 90% height = 60%/>
.caption[Hough Transform with Theta Filtering Hamby 252 Bullet 1 Land 3]
]

---
class: primary
# Hough Transform

.center[
<img src = "charlotte/06_10_2019_Update/Hamby_252_Bullet1_Land3_BestFit.png" width = 90% height = 60%/>
.caption[Hough line closest to middle two-thirds of bullet]
]

---
class: primary
# Current status

- Hough groove method is currently a part of bulletxtrctr and grooveFinder

- Changing the `get_grooves_hough` function to now return two functions that calculate the index of each left and right groove location based on the height at which a cross cut is taken

---
class: primary
# Next steps:

- Exploring Hough transform options when grooves are excluded
  
  - Also trying to write up some of this work in a coherent manner
  
  - fine-tuning of parameters, comparison with other methods

---
class: inverse
# Kiegan

---
class: primary   
# Project Background
 
My responsibilities on the bullet project:

- Automated groove ID methods  
    - **Publishing work**  
- Scanning Variability Study  
    - Pilot study data analysis  
    - Study design and implementation  
    - **Model development**  
    - **Data analysis**  
    - **Publishing work**

General update:

- Worked booth and gave a talk at AFTE conference
- Lots of good feedback from people  
- Interest in training for FA/TM examiners