EURASIP Journal on Image and Video Processing

Table 4 Accuracy of human posture recognition

From: Hybrid approach for human posture recognition using anthropometry and BP neural network based on Kinect V2

	A	B	C	D	E	F	Average
Set 1	600/600	544/560	391/400	480/480	480/480	400/400	2895/2920
Acc1	100%	97.14%	97.75%	100%	100%	100%	99.14%
Set 2	780/780	719/728	505/520	624/624	621/624	511/520	3760/3796
Acc 2	100%	98.76%	97.12%	100%	99.52%	98.27%	99.05%
Set 3	1380/1380	1263/1288	896/920	1104/1104	1101/1104	911/920	6655/6716
Acc 3	100%	98.06%	97.39%	100%	99.73%	99.02%	99.09%

A denotes standing, B denotes sitting; C denotes kneeling; D denotes sitting cross-legged; E denotes lying; and F denotes bending; the average represents the average precision of six postures which are made in one experiment; Acc_i (i = 1, 2, 3) represents the accuracy of one pose in the Set_i group

Back to article page