Skip to main content
. 2020 Apr 15;20(8):2226. doi: 10.3390/s20082226

Table 2.

A comparison between scenarios in which (i) all the original frames in the specified motion are utilized in the process of action recognition, (ii) only the keyframes extracted through Algorithm 1 are employed in the process of action recognition, (iii) ξf keyframes selected randomly are used, and (iv) ξf keyframes selected randomly are used.

Motions (i) All Frames (ii) Keyframes (iii) (ξf) (iv) (ξf)
Frames Acc. Frames Acc. Frames Acc. Frames Acc.
standUpLie 327 50% 93 100% 93 40% 186 50%
lieDownFloor 277 50% 81 100% 81 50% 162 50%
throwStandingHighR 239 66.66% 98 100% 98 69.97% 196 66.66%
standUpSitChair 172 50% 31 100% 31 55% 62 50%
grabMiddleR 151 66.66% 35 100% 35 59.99% 70 69.94%
depositMiddleR 142 100% 20 100% 20 70% 40 100%
sitDownTable 117 50% 11 100% 11 55% 22 40%
standUpSitTable 177 100% 24 100% 24 90% 48 80%
grabLowR 186 66.66% 47 66.66% 47 43.32% 94 39.97%
turnLeft 145 100% 29 100% 29 79.99% 58 100%
punchLFront 226 100% 77 100% 77 85% 154 97.75%
sitDownFloor 159 100% 54 100% 54 75% 108 100%
depositLowR 224 33.33% 52 33.33% 52 33.33% 104 33.33%
squat 536 100% 89 100% 89 98.57% 178 100%
elbowToKnee 409 100% 162 100% 162 100% 324 100%