SmarterCanes

The Dataset

IMU recordings from an instrumented walking cane, and the metadata that labels them. Consumed by ../Analyzer — see Analyzer/README.md for the commands that read this data.

Layout

data/ — every rec_*.csv recording, flat, no subfolders
split.csv — filename,split table labelling each recording train, test or walk
description.csv — semicolon-separated anomaly type per recording number, for the ae-anomaly report only; nothing in the pipeline branches on it

CSV format

A session is stored as one or more 30-second segment files, rec_XXXXX_segNNN_annY.csv, where XXXXX is the session index, NNN the zero-based segment number, and Y ∈ {-1, 0, 1} the stop annotation (shared by every segment of the session). t_ms restarts from 0 in each segment; consecutive segments of one session are read back-to-back as a single continuous recording by Analyzer/utils/io.py’s load_session.

t_ms,ax_mg,ay_mg,az_mg,gx_mdps,gy_mdps,gz_mdps,dist_mm,event
0,100,50,980,10,20,5,-1,0
10,102,48,982,12,18,6,842.5,0
20,101,49,981,11,19,5,843.1,1

t_ms — milliseconds since the segment started
ax/ay/az — acceleration in millig (loaded as g)
gx/gy/gz — rotation in millidegrees/sec (loaded as deg/s)
dist_mm — distance in millimetres from the Modulino Distance sensor; -1 if no reading is available yet
event — 1 on the first row sampled after a single click during recording, 0 otherwise

dist_mm and event were added by later firmware revisions, so a recording may carry either, both or neither; Analyzer/utils/io.py resolves columns by name and exposes has_dist / has_event.

The `annY` annotation

Y is a per-session label baked into every segment’s filename:

annotation	meaning
`ann0`	normal gait throughout
`ann1`	abnormal gait throughout
`ann-1`	normal, except at least one marked moment (found via the in-recording `event` column)
none	legacy recording, predates the annotation — treated as normal

`split.csv`

One row per file in data/, assigning it to exactly one of:

train (23 files) — normal gait only (ann0). What ae-anomaly fits its alignment template, normalization statistics, model weights and decision threshold on.
test (27 files) — labelled, normal and anomalous alike. What ae-anomaly scores, frozen, with description.csv supplying the anomaly type per session for its report. Includes rec_00017 (ann0, the held-out normal control session) alongside rec_00018–rec_00028 (ann1/ann-1, the labelled anomalies).
walk (32 files) — the rest of the pool: ordinary walking with no role in the ae-anomaly protocol.

Analyzer’s exploratory commands (step-count, step-accuracy, step-ae) take a --split flag over this table, but default to neither a single label nor the whole file: their default, normal, is every ann0/legacy recording regardless of label — train, walk, and the ann0 files inside test — since ordinary gait analysis has no use for the labelled anomalies ae-anomaly exists to score. all opts into the whole pool, anomalies included; a comma-separated subset of train, test, walk selects by label directly.