Heiko Dankert's Methods Paper 2008

2. Experimental Setup

2.1 Arena and fly handling

• Geometry, construction, fluon, food, introducing the flies and letting them settle

2.2 Camera and lighting

(preferred implementation and good ranges)

• Camera positioning,
• position of the lights, brightness in lumens (use your camera's meter, that is good enough)
• lens,
• resolution (pixels per fly), quality of hte images

2.3 Computing

• Acquisition card,
• data rates,
• storage,
• PC GHz,
• compression codecs

2.4 User interface

Here we have a user's point of view of the software. It allows you to describe what is the input and what is the output of each software module.

• Loading movies into tracker, spitting out the coordinates
• Loading the coordinates and spitting out the events

3. Tracking and feature extraction

3.1 Detecting the fly

• Gaussian model of brightness for fly and background
• Setting the thresholds
• Handling multiple flies in the same blob

3.2 Measuring the fly's position and orientation

• Fitting ellipse: measuring x, y, theta, A (from ellipse and from pixel count)
• Detecting dot on fly
• Detecting wings
• Head-tail discrimination
• Assessing the amount of measurement noise
• Nearest-neighbor trajectory connection (and what you do when this fails)

3.3 Sources of errors

• Reflections on arena's walls
• Flies walking on boundaries
• Flying flies
• Jumps
• Blur

4. Behavior detection and classification

4.1 Features

• Primary features: x,y,theta,A,wings
• Derived features

4.2 Classification

• K-nn classifier
• Acquiring ground truth

4.3 Behaviors

• Lunges
• Tussling
• ...

5. Validation

• Describe ground truth
• Show ROC