Posture Model Data Handling Report 12/02/2025

Input: Radar CSV Data Row

Frame Count, TrackID, posy, posz, vely, accy, accz, pointy0, pointz0, snr0, … , pointy49, pointz49, snr49

Step 1: Ignore frame if number of points < 4

Step 2: Insert Classification Column, Drop Frame Count Column

Classification, TrackID, posy, posz, vely, accy, accz, pointy0, pointz0, snr0, … , pointy49, pointz49, snr49

Step 3: Convert into frame with only 4 points (2 max height, 2 min height)

Classification, posy, posz, vely, velz, accy, accz, y0, z0, snr0, y1, z1, snr1, y2, z2, snr2, y3, z3, snr3

Step 4: Filter out frame if values are out of bounds (Optional)

Step 5: Sort into seperate dataframes for training and remove classification

Output: Dataframe ready to be used in training

posy, posz, vely, velz, accy, accz, y0, z0, snr0, y1, z1, snr1, y2, z2, snr2, y3, z3, snr3

Notes:

• Number of points in each training row is currently 4 due to low average number of points per input row. Can be changed to be higher later if sensitivity is increased
• Filtering out rows with extreme values is implemented for viewing the data easier and training on clean data. Not necessarily required for final implementation.
• Classification column is inserted with the correct classification based on which folder the data is stored in.

Relative Posture Model Data Handling Report 26/03/2025

Input: Radar CSV Data Row

Frame Count, TrackID, posy, posz, vely, accy, accz, pointy0, pointz0, snr0, … , pointy49, pointz49, snr49

Step 1: Ignore frame if number of points < 6

Step 2: Insert Classification Column, Drop Frame Count Column

Classification, TrackID, posy, posz, vely, accy, accz, pointy0, pointz0, snr0, … , pointy49, pointz49, snr49

Step 3: Convert into frame with only 6 points, where each y point is yn - posy (3 max height, 3 min height)

Classification, posy, posz, vely, velz, accy, accz, y0, z0, snr0, y1, z1, snr1, y2, z2, snr2, y3, z3, snr3, y4, z4, snr4, y5, z5, snr5

Step 4: Filter out frame if values are out of bounds (Optional)

Step 5: Sort into seperate dataframes for training and remove classification

Output: Dataframe ready to be used in training

posy, posz, vely, velz, accy, accz, y0, z0, snr0, y1, z1, snr1, y2, z2, snr2, y3, z3, snr3, y4, z4, snr4, y5, z5, snr5

Notes:

• Number of points in each training row is currently 4 due to low average number of points per input row. Can be changed to be higher later if sensitivity is increased
• Filtering out rows with extreme values is implemented for viewing the data easier and training on clean data. Not necessarily required for final implementation.
• Classification column is inserted with the correct classification based on which folder the data is stored in.