Data Collection Protocol

[charlieforward9]

This protocol updates this as the new data collection protocol for the binary pose classification model (with considerations to easily extend into multiclass)

Collection

Instruct the patient

Stand still with your arms crossed for about 3 seconds. Then, simulate a fall and recovery for about 3 seconds. Repeat this loop 5 times.

Conditions for accurate detections

Proximity: Patients' landmark boundaries must be as close to the fixed bounding box as possible.
Speed: The patient needs to be moving at a reasonable rate.
Stability: Camera must be static and level.
Environment: Camera positioning and ambient lighting must make the subject clearly visible.

Storage

• PNG image of the patient
• CSV file of the image metadata and the detection
• Metadata (CSV): imageHeight, imageWidth, landmarkRatio, classification_placeholder <0/1> (see Validation Section)
• Detection: [33,5] Pose Output

Validation

Each image will be labeled in the validation interface based on this map

All the data (minus invalid) will be stored in the 'valid' folder in the database and manually sent to @DrBubbles42 for model training.

@Timothyrichpt @BrodyGalloway