Step counting is incorrect in optimisation/stats.js

[AntonIsraelsson]

The program counts each state change and not the number of times a person sets down their foot.

Therefore it counts two steps when we only put the foot down once:

State	behavior	expected
...
10
00	steps++
01	steps++	steps ++

And doesn't recognize that a step has taken place when this happens: ...	State	behavior	expected
...
10
01		steps++

---

Changing the for loop to the following seems to fix this problem.

for (let i = 0; i < lines.length; i++) {
  // 578088453544940,1,1
  let line = lines[i]
  let cols = line.split(',')
  if (cols.length > 1) {
    let time = parseInt(cols[0])
    let foot1 = parseInt(cols[1])
    let foot2 = parseInt(cols[2])
    if (i === 0) {
      start = time
      prevfoot1 = foot1
      prevfoot2 = foot2
    } else duration = time - start
    // If foot1 goes from 0 to 1:
    if (foot1 - prevfoot1 === 1) {
      steps++
    }
    // If foot2 goes from 0 to 1:
    if (foot2 - prevfoot2 === 1) {
      steps++
    }
    prevfoot1 = foot1
    prevfoot2 = foot2
  }
}

[dariosalvi78]

so you mean that it counts one extra step for the whole test?

[AntonIsraelsson]

No, it seems to be off by a factor of around 2

[dariosalvi78]

yes well, right and left step, of course.. To be honest, on longer distances with hundreds of steps it's not a big deal, but it would be good to fix anyhow. Maybe you want to send a pull request?

[AntonIsraelsson]

To be clear it's off by over a hundred steps for each dataset. A frequency of 4 steps per second is unreasonable for most kinds of walks.

Steps Original	Steps New	Duration (s)	Steps/s, Original	Steps/s, New
398	198	107	3.7	1.9
455	248	132	3.4	1.9
475	243	137	3.5	1.8
535	270	138	3.9	2.0
427	261	138	3.1	1.9
516	263	139	3.7	1.9
517	260	142	3.6	1.8
557	284	143	3.9	2.0
380	195	144	2.6	1.4
411	206	144	2.9	1.4
523	275	144	3.6	1.9
514	272	144	3.6	1.9
537	277	144	3.7	1.9
532	266	145	3.7	1.8
529	265	145	3.6	1.8
533	267	146	3.7	1.8
497	255	147	3.4	1.7
468	280	147	3.2	1.9
505	257	147	3.4	1.7
493	285	148	3.3	1.9
533	268	151	3.5	1.8
485	282	151	3.2	1.9
605	305	152	4.0	2.0
611	306	152	4.0	2.0
517	278	152	3.4	1.8
533	266	152	3.5	1.8
581	295	153	3.8	1.9
643	322	159	4.0	2.0
646	323	162	4.0	2.0
565	288	162	3.5	1.8
582	292	163	3.6	1.8
585	302	164	3.6	1.8
473	289	165	2.9	1.8
547	312	168	3.3	1.9
571	302	169	3.4	1.8
575	299	169	3.4	1.8
537	303	171	3.1	1.8
577	311	173	3.3	1.8
515	330	176	2.9	1.9
557	316	177	3.1	1.8
528	306	180	2.9	1.7
465	314	187	2.5	1.7
693	362	189	3.7	1.9
370	277	189	2.0	1.5
707	364	191	3.7	1.9
431	301	192	2.2	1.6
502	335	208	2.4	1.6
463	365	219	2.1	1.7

[dariosalvi78]

I don't get it, what is off? the ground truth or the estimated steps from the algorithm?

[AntonIsraelsson]

The ground truth

[dariosalvi78]

I am very surprised, we validated the GT device with lab measurements. This is how the steps are counted in the app. See from line 239. It looks like it's not double counting.

I think we used those steps as GT in the optimization process, not the ones computed in stats.js. Actually I cannot remember what stats.js was used for. It was some time ago!

[AntonIsraelsson]

It might just be an error with the stats.js file, as all other file-steps seem to be correct.

How was it validated?

[dariosalvi78]

we validated the GT app by comparing the N of steps computed there with what was counted manually (2 separate people observed the test) over a short walk. I think the issue is only with the stats.js file, which I am not sure it was actually used for anything, so hopefully our results are not affected by it. In any case thanks for finding and correcting the issue!

If you need any help you can also write me an email at dario.salvi.work at gmail