-
Notifications
You must be signed in to change notification settings - Fork 0
/
myHandMoveRecognition.py
141 lines (129 loc) · 8.9 KB
/
myHandMoveRecognition.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
from myGlobalVariables import *
from sklearn.ensemble import IsolationForest
# 手部运动识别
# 输入手部轨迹列表handTrack，该列表形如[(x1,y1),(x2,y2),...]，连续记录过去若干帧的手部位置
# hand是可选输入参数，主要是调试时用于在画面绘制一些图形用，使程序计算结果更直观，便于调试
# 返回手部运动方向(left, right,up, down, clockwise, anticlockwise, static)及置信度
def handMovRecog(handTrack, hand=None):
# 测试用轨迹数据，17点，第2、14、16号数据为离群点
# handTrack = [(325, 306), (360, 314), (567, 426), (367, 326), (352, 330), (333, 326), (328, 317), (327, 305),
# (328, 289), (343, 280), (362, 274), (380, 273), (398, 271), (400, 272), (534, 215), (303, 303),
# (-1, -1)]
# print('handTrack:', len(handTrack), handTrack)
handMovement = "unkown" # 根据轨迹判断结果，用运动方向名称给此变量赋值
confidence = 1 # 根据轨迹判断结果的自信程度设置此值
maxDisError = 100 # 容许的两点之间各维度最大距离误差，用于清理离群点
minRadius = 20 # 圆周运动最小识别半径
minDisT = 50 # 容许的轨迹点间最大距离的最小平均值，超过此平均值阈值才视为运动
maxDist = 100 # 直线运动时，与运动方向垂直方向上的波动范围不能超过此值
global TrackingHand # 手部跟踪器
handTrack = [(x, y) for x, y in handTrack if x > 0] # 滤掉初始值(0,0)和未分割到手势值(-1,-1)
if len(handTrack) > 0:
# 利用iForest算法计算轨迹中的离群点
iForest = IsolationForest(n_estimators=10)
iForest.fit(handTrack)
Outliers = iForest.predict(handTrack) # 1为正常值，-1为异常值
# print(Outliers)
# handTrackFiltered=[handTrack[i] for i in range(len(handTrack)) if Outliers[i] == 1] # 滤掉异常值
if Outliers[0] == 1: # 处理起点，不是离群点就留下
handTrackFiltered = [handTrack[0]]
else: # 是离群点就舍弃
handTrackFiltered = []
# 再次检测离群点，算法算出的离群点不都准，偏离未超过maxDisError的“离群点”也作为正常点保留。未考虑子序列离群。
for i in range(1, len(handTrack) - 1):
if Outliers[i] == -1:
d1 = (handTrack[i][0] - handTrack[i - 1][0]) ** 2 + (
handTrack[i][1] - handTrack[i - 1][1]) ** 2 # 与前一点的平方距离
d2 = (handTrack[i][0] - handTrack[i + 1][0]) ** 2 + (
handTrack[i][1] - handTrack[i + 1][1]) ** 2 # 与后一点的平方距离
if d1 < maxDisError ** 2 and d2 < maxDisError ** 2: # 偏差
Outliers[i] = 1 # 偏差不大的点重新置为正常点
handTrackFiltered.append(handTrack[i])
else:
handTrackFiltered.append(handTrack[i])
if Outliers[-1] == 1: # 处理终点，不是离群点就留下
handTrackFiltered.append(handTrack[-1])
# print('handTrackFiltered:', len(handTrackFiltered), handTrackFiltered)
if len(handTrackFiltered) > 1:
# 计算有效轨迹点之间最大距离
minXY = np.min(handTrackFiltered, axis=0)
maxXY = np.max(handTrackFiltered, axis=0)
dXY = maxXY - minXY
# print(dXY)
# 计算相邻点间各维度差
dxdy = []
for i in range(1, len(handTrackFiltered)):
dx = handTrackFiltered[i][0] - handTrackFiltered[i - 1][0]
dy = handTrackFiltered[i][1] - handTrackFiltered[i - 1][1]
dxdy.append((dx, dy))
dxdymean = np.mean(dxdy, axis=0) # 计算均值，用于判断方向
# dxdyabsmean = np.mean(np.abs(dxdy), axis=0) # 计算绝对值均值，用于判断偏差
TrackArea = cv2.contourArea(np.array(handTrackFiltered)) # 轨迹面积
TrackPerimeter = cv2.arcLength(np.array(handTrackFiltered), True) # 轨迹周长
# TrackRadius = TrackPerimeter/np.pi/2 # 估算圆形轨迹半径
# 圆形度=（4π * 面积） / （周长 * 周长）
TrackCircularity = 0.0
if TrackPerimeter > 0:
TrackCircularity = 4 * np.pi * TrackArea / (TrackPerimeter * TrackPerimeter)
# print('圆度：',TrackCircularity,'水平跨度：',dXY[0],'垂直跨度：',dXY[1],'方向：',dxdymean)
if dXY[0] > minRadius * 2 and dXY[1] > minRadius * 2 and TrackCircularity > 0.4: # 纵横运动距离均大于半径阈值的2倍，可能为圆形运动
# 判断旋转方向的思路：在图像坐标系中（y轴向下），顺时针旋转，方位角（“轨迹圆心-手”向量与x轴夹角）增大
# 统计轨迹点坐标均值可以计算出轨迹圆的中心，利用反正切函数可以算出方位角（要折算到0-2pi之间，过x正半轴需做特殊考虑）
# 统计方位角增量，可以判定顺时针还是逆时针
circleCenter = np.mean(handTrackFiltered, axis=0) # 计算圆形运动轨迹中心
if hand is not None: # 如果可选参数传入了hand，就画出来中间计算结果
cv2.circle(hand, (int(circleCenter[0]), int(circleCenter[1])), 10, 255, -1)
cv2.line(hand, (int(circleCenter[0]), int(circleCenter[1])),
(handTrackFiltered[len(handTrackFiltered) - 1]), 255, 1)
lastAngle = 10 # 记录上一个轨迹点方位角，随便设置一个2pi范围以外的初值
DeltaAngle = [] # 记录手挥过的角度差
for handPos in handTrackFiltered:
aTan = np.arctan2(handPos[1] - circleCenter[1], handPos[0] - circleCenter[0]) # 计算方位角
if aTan < 0: # 转换到0-2pi之间
aTan = aTan + np.pi * 2
if lastAngle != 10: # 如果不是初值，那就是赋过值
if (aTan < np.pi / 2 and lastAngle > np.pi * 1.5):
DeltaAngle.append(aTan) # 计算方位角增量，顺时针转回一圈时需做特殊判断
elif (lastAngle < np.pi / 2 and aTan > np.pi * 1.5):
DeltaAngle.append(0 - lastAngle) # 计算方位角增量，逆时针转回一圈时需做特殊判断
else:
DeltaAngle.append(aTan - lastAngle)
lastAngle = aTan
if len(DeltaAngle)>0:
dAngleMean = np.mean(DeltaAngle) # 统计方位角增量正负
else:
dAngleMean=0
if dAngleMean < 0:
handMovement = "anticlockwise"
else:
handMovement = "clockwise"
confidence = 1 - TrackCircularity * 0.1
elif dXY[0] > dXY[1] and dXY[0] > minDisT and dXY[1] < maxDist: # 横向运动幅度大于纵向且大于运动阈值，纵向运动小于波动阈值
if not TrackingHand is None: # 利用Kalman滤波器预测位置与当前位置关系判断运动方向
if TrackingHand.prediction[0] > TrackingHand.measurement[0]:
handMovement = "right"
elif TrackingHand.prediction[0] < TrackingHand.measurement[0]:
handMovement = "left"
else: # 旧方法
if dxdymean[0] > 0: # 向x轴正向运动
handMovement = "right"
elif dxdymean[0] < 0: # 向x轴负向运动
handMovement = "left"
confidence = 1 - dXY[1] / dXY[0] * 0.1
elif dXY[1] > dXY[0] and dXY[1] > minDisT and dXY[0] < maxDist: # 纵向运动幅度大于横向且大于运动阈值，横向运动小于波动阈值
if not TrackingHand is None: # 利用Kalman滤波器预测位置与当前位置关系判断运动方向
if TrackingHand.prediction[1] > TrackingHand.measurement[1]:
handMovement = "down"
elif TrackingHand.prediction[1] < TrackingHand.measurement[1]:
handMovement = "up"
else: # 旧方法
if dxdymean[1] > 0: # 向y轴正向运动
handMovement = "down"
elif dxdymean[1] < 0: # 向y轴负向运动
handMovement = "up"
confidence = 1 - dXY[0] / dXY[1] * 0.1
elif dXY[1] < minDisT and dXY[0] < minDisT: # 横向纵向偏差均小于运动阈值
handMovement = "static"
confidence = 1 - (dXY[0] / minDisT) * (dXY[0] / minDisT) * 0.1
# print(handMovement)
return handMovement, confidence