当检测到某个手势时播放.wav文件，
软硬件版本信息
k230 Micropython
错误日志
没办法播放声音只能检测手势
尝试解决过程
我根据论坛的回答也采用了两个线程，但是还是不能播放
补充材料
,,
2 Answers
请修改代码为markdown格式，这样便于快速复现问题。
===================2024/11/21========================
示例如下：
from libs.PipeLine import PipeLine, ScopedTiming
from libs.AIBase import AIBase
from libs.AI2D import Ai2d
import os
import ujson
from media.media import * # 导入media模块，用于初始化vb buffer
from media.pyaudio import * # 音频模块
import media.wave as wave # 导入wav模块，用于保存和加载wav音频文件
import nncase_runtime as nn
import ulab.numpy as np
import image
import aicube
import random
import gc
import sys
import _thread
# 全局变量，用于存储最新的手势识别结果
global latest_gesture
latest_gesture = None
audio_run=True
# 自定义手掌检测任务类
class HandDetApp(AIBase):
def __init__(self,kmodel_path,labels,model_input_size,anchors,confidence_threshold=0.2,nms_threshold=0.5,nms_option=False, strides=[8,16,32],rgb888p_size=[1920,1080],display_size=[1920,1080],debug_mode=0):
super().__init__(kmodel_path,model_input_size,rgb888p_size,debug_mode)
# kmodel路径
self.kmodel_path=kmodel_path
self.labels=labels
# 检测模型输入分辨率
self.model_input_size=model_input_size
# 置信度阈值
self.confidence_threshold=confidence_threshold
# nms阈值
self.nms_threshold=nms_threshold
# 锚框,目标检测任务使用
self.anchors=anchors
# 特征下采样倍数
self.strides = strides
# NMS选项，如果为True做类间NMS,如果为False做类内NMS
self.nms_option = nms_option
# sensor给到AI的图像分辨率，宽16字节对齐
self.rgb888p_size=[ALIGN_UP(rgb888p_size[0],16),rgb888p_size[1]]
# 视频输出VO分辨率，宽16字节对齐
self.display_size=[ALIGN_UP(display_size[0],16),display_size[1]]
# debug模式
self.debug_mode=debug_mode
# Ai2d实例用于实现预处理
self.ai2d=Ai2d(debug_mode)
# 设置ai2d的输入输出的格式和数据类型
self.ai2d.set_ai2d_dtype(nn.ai2d_format.NCHW_FMT,nn.ai2d_format.NCHW_FMT,np.uint8, np.uint8)
# 配置预处理操作，这里使用了pad和resize，Ai2d支持crop/shift/pad/resize/affine，具体代码请打开/sdcard/app/libs/AI2D.py查看
def config_preprocess(self,input_image_size=None):
with ScopedTiming("set preprocess config",self.debug_mode > 0):
# 初始化ai2d预处理配置，默认为sensor给到AI的尺寸，可以通过设置input_image_size自行修改输入尺寸
ai2d_input_size = input_image_size if input_image_size else self.rgb888p_size
# 计算padding参数并应用pad操作，以确保输入图像尺寸与模型输入尺寸匹配
top, bottom, left, right = self.get_padding_param()
self.ai2d.pad([0, 0, 0, 0, top, bottom, left, right], 0, [114, 114, 114])
# 使用双线性插值进行resize操作，调整图像尺寸以符合模型输入要求
self.ai2d.resize(nn.interp_method.tf_bilinear, nn.interp_mode.half_pixel)
# 构建预处理流程,参数为预处理输入tensor的shape和预处理输出的tensor的shape
self.ai2d.build([1,3,ai2d_input_size[1],ai2d_input_size[0]],[1,3,self.model_input_size[1],self.model_input_size[0]])
# 自定义当前任务的后处理，用于处理模型输出结果，这里使用了aicube库的anchorbasedet_post_process接口
def postprocess(self,results):
with ScopedTiming("postprocess",self.debug_mode > 0):
dets = aicube.anchorbasedet_post_process(results[0], results[1], results[2], self.model_input_size, self.rgb888p_size, self.strides, len(self.labels), self.confidence_threshold, self.nms_threshold, self.anchors, self.nms_option)
# 返回手掌检测结果
return dets
# 计算padding参数，确保输入图像尺寸与模型输入尺寸匹配
def get_padding_param(self):
# 根据目标宽度和高度计算比例因子
dst_w = self.model_input_size[0]
dst_h = self.model_input_size[1]
input_width = self.rgb888p_size[0]
input_high = self.rgb888p_size[1]
ratio_w = dst_w / input_width
ratio_h = dst_h / input_high
# 选择较小的比例因子，以确保图像内容完整
if ratio_w < ratio_h:
ratio = ratio_w
else:
ratio = ratio_h
# 计算新的宽度和高度
new_w = int(ratio * input_width)
new_h = int(ratio * input_high)
# 计算宽度和高度的差值，并确定padding的位置
dw = (dst_w - new_w) / 2
dh = (dst_h - new_h) / 2
top = int(round(dh - 0.1))
bottom = int(round(dh + 0.1))
left = int(round(dw - 0.1))
right = int(round(dw + 0.1))
return top, bottom, left, right
# 自定义手势关键点分类任务类
class HandKPClassApp(AIBase):
def __init__(self,kmodel_path,model_input_size,rgb888p_size=[1920,1080],display_size=[1920,1080],debug_mode=0):
super().__init__(kmodel_path,model_input_size,rgb888p_size,debug_mode)
# kmodel路径
self.kmodel_path=kmodel_path
# 检测模型输入分辨率
self.model_input_size=model_input_size
# sensor给到AI的图像分辨率，宽16字节对齐
self.rgb888p_size=[ALIGN_UP(rgb888p_size[0],16),rgb888p_size[1]]
# 视频输出VO分辨率，宽16字节对齐
self.display_size=[ALIGN_UP(display_size[0],16),display_size[1]]
self.crop_params=[]
# debug模式
self.debug_mode=debug_mode
# Ai2d实例用于实现预处理
self.ai2d=Ai2d(debug_mode)
# 设置ai2d的输入输出的格式和数据类型
self.ai2d.set_ai2d_dtype(nn.ai2d_format.NCHW_FMT,nn.ai2d_format.NCHW_FMT,np.uint8, np.uint8)
# 配置预处理操作，这里使用了crop和resize，Ai2d支持crop/shift/pad/resize/affine，具体代码请打开/sdcard/app/libs/AI2D.py查看
def config_preprocess(self,det,input_image_size=None):
with ScopedTiming("set preprocess config",self.debug_mode > 0):
ai2d_input_size=input_image_size if input_image_size else self.rgb888p_size
self.crop_params = self.get_crop_param(det)
self.ai2d.crop(self.crop_params[0],self.crop_params[1],self.crop_params[2],self.crop_params[3])
self.ai2d.resize(nn.interp_method.tf_bilinear, nn.interp_mode.half_pixel)
self.ai2d.build([1,3,ai2d_input_size[1],ai2d_input_size[0]],[1,3,self.model_input_size[1],self.model_input_size[0]])
# 自定义后处理，得到手掌手势结果和手掌关键点数据
def postprocess(self,results):
with ScopedTiming("postprocess",self.debug_mode > 0):
results=results[0].reshape(results[0].shape[0]*results[0].shape[1])
results_show = np.zeros(results.shape,dtype=np.int16)
results_show[0::2] = results[0::2] * self.crop_params[3] + self.crop_params[0]
results_show[1::2] = results[1::2] * self.crop_params[2] + self.crop_params[1]
gesture=self.hk_gesture(results_show)
results_show[0::2] = results_show[0::2] * (self.display_size[0] / self.rgb888p_size[0])
results_show[1::2] = results_show[1::2] * (self.display_size[1] / self.rgb888p_size[1])
return results_show,gesture
# 计算crop参数
def get_crop_param(self,det_box):
x1, y1, x2, y2 = det_box[2],det_box[3],det_box[4],det_box[5]
w,h= int(x2 - x1),int(y2 - y1)
w_det = int(float(x2 - x1) * self.display_size[0] // self.rgb888p_size[0])
h_det = int(float(y2 - y1) * self.display_size[1] // self.rgb888p_size[1])
x_det = int(x1*self.display_size[0] // self.rgb888p_size[0])
y_det = int(y1*self.display_size[1] // self.rgb888p_size[1])
length = max(w, h)/2
cx = (x1+x2)/2
cy = (y1+y2)/2
ratio_num = 1.26*length
x1_kp = int(max(0,cx-ratio_num))
y1_kp = int(max(0,cy-ratio_num))
x2_kp = int(min(self.rgb888p_size[0]-1, cx+ratio_num))
y2_kp = int(min(self.rgb888p_size[1]-1, cy+ratio_num))
w_kp = int(x2_kp - x1_kp + 1)
h_kp = int(y2_kp - y1_kp + 1)
return [x1_kp, y1_kp, w_kp, h_kp]
# 求两个vector之间的夹角
def hk_vector_2d_angle(self,v1,v2):
with ScopedTiming("hk_vector_2d_angle",self.debug_mode > 0):
v1_x,v1_y,v2_x,v2_y = v1[0],v1[1],v2[0],v2[1]
v1_norm = np.sqrt(v1_x * v1_x+ v1_y * v1_y)
v2_norm = np.sqrt(v2_x * v2_x + v2_y * v2_y)
dot_product = v1_x * v2_x + v1_y * v2_y
cos_angle = dot_product/(v1_norm*v2_norm)
angle = np.acos(cos_angle)*180/np.pi
return angle
# 根据手掌关键点检测结果判断手势类别
def hk_gesture(self,results):
with ScopedTiming("hk_gesture",self.debug_mode > 0):
angle_list = []
for i in range(5):
angle = self.hk_vector_2d_angle([(results[0]-results[i*8+4]), (results[1]-results[i*8+5])],[(results[i*8+6]-results[i*8+8]),(results[i*8+7]-results[i*8+9])])
angle_list.append(angle)
thr_angle,thr_angle_thumb,thr_angle_s,gesture_str = 65.,53.,49.,None
if 65535. not in angle_list:
if (angle_list[0]>thr_angle_thumb) and (angle_list[1]>thr_angle) and (angle_list[2]>thr_angle) and (angle_list[3]>thr_angle) and (angle_list[4]>thr_angle):
gesture_str = "fist"
elif (angle_list[0]<thr_angle_s) and (angle_list[1]<thr_angle_s) and (angle_list[2]<thr_angle_s) and (angle_list[3]<thr_angle_s) and (angle_list[4]<thr_angle_s):
gesture_str = "five"
elif (angle_list[0]<thr_angle_s) and (angle_list[1]<thr_angle_s) and (angle_list[2]>thr_angle) and (angle_list[3]>thr_angle) and (angle_list[4]>thr_angle):
gesture_str = "gun"
elif (angle_list[0]<thr_angle_s) and (angle_list[1]<thr_angle_s) and (angle_list[2]>thr_angle) and (angle_list[3]>thr_angle) and (angle_list[4]<thr_angle_s):
gesture_str = "love"
elif (angle_list[0]>5) and (angle_list[1]<thr_angle_s) and (angle_list[2]>thr_angle) and (angle_list[3]>thr_angle) and (angle_list[4]>thr_angle):
gesture_str = "one"
elif (angle_list[0]<thr_angle_s) and (angle_list[1]>thr_angle) and (angle_list[2]>thr_angle) and (angle_list[3]>thr_angle) and (angle_list[4]<thr_angle_s):
gesture_str = "six"
elif (angle_list[0]>thr_angle_thumb) and (angle_list[1]<thr_angle_s) and (angle_list[2]<thr_angle_s) and (angle_list[3]<thr_angle_s) and (angle_list[4]>thr_angle):
gesture_str = "three"
elif (angle_list[0]<thr_angle_s) and (angle_list[1]>thr_angle) and (angle_list[2]>thr_angle) and (angle_list[3]>thr_angle) and (angle_list[4]>thr_angle):
gesture_str = "thumbUp"
elif (angle_list[0]>thr_angle_thumb) and (angle_list[1]<thr_angle_s) and (angle_list[2]<thr_angle_s) and (angle_list[3]>thr_angle) and (angle_list[4]>thr_angle):
gesture_str = "yeah"
return gesture_str
# 手掌关键点分类任务
class HandKeyPointClass:
def __init__(self,hand_det_kmodel,hand_kp_kmodel,det_input_size,kp_input_size,labels,anchors,confidence_threshold=0.25,nms_threshold=0.3,nms_option=False,strides=[8,16,32],rgb888p_size=[1280,720],display_size=[1920,1080],debug_mode=0):
# 手掌检测模型路径
self.hand_det_kmodel=hand_det_kmodel
# 手掌关键点模型路径
self.hand_kp_kmodel=hand_kp_kmodel
# 手掌检测模型输入分辨率
self.det_input_size=det_input_size
# 手掌关键点模型输入分辨率
self.kp_input_size=kp_input_size
self.labels=labels
# anchors
self.anchors=anchors
# 置信度阈值
self.confidence_threshold=confidence_threshold
# nms阈值
self.nms_threshold=nms_threshold
self.nms_option=nms_option
self.strides=strides
# sensor给到AI的图像分辨率，宽16字节对齐
self.rgb888p_size=[ALIGN_UP(rgb888p_size[0],16),rgb888p_size[1]]
# 视频输出VO分辨率，宽16字节对齐
self.display_size=[ALIGN_UP(display_size[0],16),display_size[1]]
# debug_mode模式
self.debug_mode=debug_mode
self.hand_det=HandDetApp(self.hand_det_kmodel,self.labels,model_input_size=self.det_input_size,anchors=self.anchors,confidence_threshold=self.confidence_threshold,nms_threshold=self.nms_threshold,nms_option=self.nms_option,strides=self.strides,rgb888p_size=self.rgb888p_size,display_size=self.display_size,debug_mode=0)
self.hand_kp=HandKPClassApp(self.hand_kp_kmodel,model_input_size=self.kp_input_size,rgb888p_size=self.rgb888p_size,display_size=self.display_size)
self.hand_det.config_preprocess()
# run函数
def run(self,input_np):
# 执行手掌检测
det_boxes=self.hand_det.run(input_np)
boxes=[]
gesture_res=[]
for det_box in det_boxes:
# 对于检测到的每一个手掌执行关键点识别
x1, y1, x2, y2 = det_box[2],det_box[3],det_box[4],det_box[5]
w,h= int(x2 - x1),int(y2 - y1)
if (h<(0.1*self.rgb888p_size[1])):
continue
if (w<(0.25*self.rgb888p_size[0]) and ((x1<(0.03*self.rgb888p_size[0])) or (x2>(0.97*self.rgb888p_size[0])))):
continue
if (w<(0.15*self.rgb888p_size[0]) and ((x1<(0.01*self.rgb888p_size[0])) or (x2>(0.99*self.rgb888p_size[0])))):
continue
self.hand_kp.config_preprocess(det_box)
results_show,gesture=self.hand_kp.run(input_np)
gesture_res.append((results_show,gesture))
boxes.append(det_box)
return boxes,gesture_res
# 绘制效果，绘制关键点、手掌检测框和识别结果
def draw_result(self,pl,dets,gesture_res):
global latest_gesture
pl.osd_img.clear()
if len(dets)>0:
for k in range(len(dets)):
det_box=dets[k]
x1, y1, x2, y2 = det_box[2],det_box[3],det_box[4],det_box[5]
w,h= int(x2 - x1),int(y2 - y1)
if (h<(0.1*self.rgb888p_size[1])):
continue
if (w<(0.25*self.rgb888p_size[0]) and ((x1<(0.03*self.rgb888p_size[0])) or (x2>(0.97*self.rgb888p_size[0])))):
continue
if (w<(0.15*self.rgb888p_size[0]) and ((x1<(0.01*self.rgb888p_size[0])) or (x2>(0.99*self.rgb888p_size[0])))):
continue
w_det = int(float(x2 - x1) * self.display_size[0] // self.rgb888p_size[0])
h_det = int(float(y2 - y1) * self.display_size[1] // self.rgb888p_size[1])
x_det = int(x1*self.display_size[0] // self.rgb888p_size[0])
y_det = int(y1*self.display_size[1] // self.rgb888p_size[1])
pl.osd_img.draw_rectangle(x_det, y_det, w_det, h_det, color=(255, 0, 255, 0), thickness = 2)
results_show=gesture_res[k][0]
for i in range(len(results_show)/2):
pl.osd_img.draw_circle(results_show[i*2], results_show[i*2+1], 1, color=(255, 0, 255, 0),fill=False)
for i in range(5):
j = i*8
if i==0:
R = 255; G = 0; B = 0
if i==1:
R = 255; G = 0; B = 255
if i==2:
R = 255; G = 255; B = 0
if i==3:
R = 0; G = 255; B = 0
if i==4:
R = 0; G = 0; B = 255
pl.osd_img.draw_line(results_show[0], results_show[1], results_show[j+2], results_show[j+3], color=(255,R,G,B), thickness = 3)
pl.osd_img.draw_line(results_show[j+2], results_show[j+3], results_show[j+4], results_show[j+5], color=(255,R,G,B), thickness = 3)
pl.osd_img.draw_line(results_show[j+4], results_show[j+5], results_show[j+6], results_show[j+7], color=(255,R,G,B), thickness = 3)
pl.osd_img.draw_line(results_show[j+6], results_show[j+7], results_show[j+8], results_show[j+9], color=(255,R,G,B), thickness = 3)
gesture_str=gesture_res[k][1]
latest_gesture=gesture_str
pl.osd_img.draw_string_advanced( x_det , y_det-50,32, " " + str(gesture_str), color=(255,0, 255, 0))
def audio_output_thread(output_stream):
global latest_gesture
CHUNK = int(0.3 * 16000) # 每次读取音频数据的帧数，设置为0.3s的帧数16000*0.3=4800
reply_wav_file = "/sdcard/examples/utils/wozai.wav" # 音频路径
while audio_run:
if latest_gesture == "fist":
wf = wave.open(reply_wav_file, "rb")
wav_data = wf.read_frames(CHUNK)
while wav_data:
output_stream.write(wav_data)
wav_data = wf.read_frames(CHUNK)
time.sleep(1) # 时间缓冲，用于播放回复声音
wf.close()
latest_gesture=""
if __name__ == "__main__":
# 其它参数
THRESH = 0.5 # 检测阈值
SAMPLE_RATE = 16000 # 采样率16000Hz,即每秒采样16000次
CHANNELS = 1 # 通道数 1为单声道，2为立体声
FORMAT = paInt16 # 音频输入输出格式 paInt16
CHUNK = int(0.3 * 16000) # 每次读取音频数据的帧数，设置为0.3s的帧数16000*0.3=4800
# 初始化音频流
p = PyAudio()
p.initialize(CHUNK)
# 显示模式，默认"hdmi",可以选择"hdmi"和"lcd"
display_mode = "lcd"
if display_mode == "hdmi":
display_size = [1920, 1080]
else:
display_size = [800, 480]
# 手掌检测模型路径
hand_det_kmodel_path = "/sdcard/examples/kmodel/hand_det.kmodel"
# 手掌关键点模型路径
hand_kp_kmodel_path = "/sdcard/examples/kmodel/handkp_det.kmodel"
# 其他参数
anchors_path = "/sdcard/examples/utils/prior_data_320.bin"
rgb888p_size = [1920, 1080]
hand_det_input_size = [512, 512]
hand_kp_input_size = [256, 256]
confidence_threshold = 0.4
nms_threshold = 0.5
labels = ["hand"]
anchors = [26, 27, 53, 52, 75, 71, 80, 99, 106, 82, 99, 134, 140, 113, 161, 172, 245, 276]
# 初始化PipeLine，只关注传给AI的图像分辨率，显示的分辨率
pl = PipeLine(rgb888p_size=rgb888p_size, display_size=display_size, display_mode=display_mode)
pl.create()
hkc = HandKeyPointClass(hand_det_kmodel_path, hand_kp_kmodel_path, det_input_size=hand_det_input_size, kp_input_size=hand_kp_input_size, labels=labels, anchors=anchors, confidence_threshold=confidence_threshold, nms_threshold=nms_threshold, nms_option=False, strides=[8, 16, 32], rgb888p_size=rgb888p_size, display_size=display_size, debug_mode=0)
clock = time.clock()
output_stream = p.open(format=FORMAT,channels=CHANNELS,rate=SAMPLE_RATE,output=True,frames_per_buffer=CHUNK)
_thread.start_new_thread(audio_output_thread,(output_stream,))
try:
while True:
os.exitpoint()
clock.tick()
img = pl.get_frame() # 获取当前帧
det_boxes, gesture_res = hkc.run(img) # 推理当前帧
hkc.draw_result(pl, det_boxes, gesture_res) # 绘制当前帧推理结果
pl.show_image() # 展示推理结果
gc.collect()
print(clock.fps()) # 打印帧率
except Exception as e:
sys.print_exception(e)
audio_run=False
finally:
audio_run=False
time.sleep(1)
output_stream.stop_stream()
output_stream.close()
hkc.hand_det.deinit()
hkc.hand_kp.deinit()
pl.destroy()
'''
实验名称：手掌关键点分类
实验平台：01Studio CanMV K230
教程：wiki.01studio.cc
'''
from libs.PipeLine import PipeLine, ScopedTiming
from libs.AIBase import AIBase
from libs.AI2D import Ai2d
import os
import ujson
from media.media import * # 导入media模块，用于初始化vb buffer
from media.pyaudio import * # 音频模块
import media.wave as wave # 导入wav模块，用于保存和加载wav音频文件
import nncase_runtime as nn
import ulab.numpy as np
import image
import aicube
import random
import gc
import sys
import _thread
# 全局变量，用于存储最新的手势识别结果
global latest_gesture
latest_gesture = None
audio_run=True
# 自定义手掌检测任务类
class HandDetApp(AIBase):
def __init__(self,kmodel_path,labels,model_input_size,anchors,confidence_threshold=0.2,nms_threshold=0.5,nms_option=False, strides=[8,16,32],rgb888p_size=[1920,1080],display_size=[1920,1080],debug_mode=0):
super().__init__(kmodel_path,model_input_size,rgb888p_size,debug_mode)
# kmodel路径
self.kmodel_path=kmodel_path
self.labels=labels
# 检测模型输入分辨率
self.model_input_size=model_input_size
# 置信度阈值
self.confidence_threshold=confidence_threshold
# nms阈值
self.nms_threshold=nms_threshold
# 锚框,目标检测任务使用
self.anchors=anchors
# 特征下采样倍数
self.strides = strides
# NMS选项，如果为True做类间NMS,如果为False做类内NMS
self.nms_option = nms_option
# sensor给到AI的图像分辨率，宽16字节对齐
self.rgb888p_size=[ALIGN_UP(rgb888p_size[0],16),rgb888p_size[1]]
# 视频输出VO分辨率，宽16字节对齐
self.display_size=[ALIGN_UP(display_size[0],16),display_size[1]]
# debug模式
self.debug_mode=debug_mode
# Ai2d实例用于实现预处理
self.ai2d=Ai2d(debug_mode)
# 设置ai2d的输入输出的格式和数据类型
self.ai2d.set_ai2d_dtype(nn.ai2d_format.NCHW_FMT,nn.ai2d_format.NCHW_FMT,np.uint8, np.uint8)
# 配置预处理操作，这里使用了pad和resize，Ai2d支持crop/shift/pad/resize/affine，具体代码请打开/sdcard/app/libs/AI2D.py查看
def config_preprocess(self,input_image_size=None):
with ScopedTiming("set preprocess config",self.debug_mode > 0):
# 初始化ai2d预处理配置，默认为sensor给到AI的尺寸，可以通过设置input_image_size自行修改输入尺寸
ai2d_input_size = input_image_size if input_image_size else self.rgb888p_size
# 计算padding参数并应用pad操作，以确保输入图像尺寸与模型输入尺寸匹配
top, bottom, left, right = self.get_padding_param()
self.ai2d.pad([0, 0, 0, 0, top, bottom, left, right], 0, [114, 114, 114])
# 使用双线性插值进行resize操作，调整图像尺寸以符合模型输入要求
self.ai2d.resize(nn.interp_method.tf_bilinear, nn.interp_mode.half_pixel)
# 构建预处理流程,参数为预处理输入tensor的shape和预处理输出的tensor的shape
self.ai2d.build([1,3,ai2d_input_size[1],ai2d_input_size[0]],[1,3,self.model_input_size[1],self.model_input_size[0]])
# 自定义当前任务的后处理，用于处理模型输出结果，这里使用了aicube库的anchorbasedet_post_process接口
def postprocess(self,results):
with ScopedTiming("postprocess",self.debug_mode > 0):
dets = aicube.anchorbasedet_post_process(results[0], results[1], results[2], self.model_input_size, self.rgb888p_size, self.strides, len(self.labels), self.confidence_threshold, self.nms_threshold, self.anchors, self.nms_option)
# 返回手掌检测结果
return dets
# 计算padding参数，确保输入图像尺寸与模型输入尺寸匹配
def get_padding_param(self):
# 根据目标宽度和高度计算比例因子
dst_w = self.model_input_size[0]
dst_h = self.model_input_size[1]
input_width = self.rgb888p_size[0]
input_high = self.rgb888p_size[1]
ratio_w = dst_w / input_width
ratio_h = dst_h / input_high
# 选择较小的比例因子，以确保图像内容完整
if ratio_w < ratio_h:
ratio = ratio_w
else:
ratio = ratio_h
# 计算新的宽度和高度
new_w = int(ratio * input_width)
new_h = int(ratio * input_high)
# 计算宽度和高度的差值，并确定padding的位置
dw = (dst_w - new_w) / 2
dh = (dst_h - new_h) / 2
top = int(round(dh - 0.1))
bottom = int(round(dh + 0.1))
left = int(round(dw - 0.1))
right = int(round(dw + 0.1))
return top, bottom, left, right
# 自定义手势关键点分类任务类
class HandKPClassApp(AIBase):
def __init__(self,kmodel_path,model_input_size,rgb888p_size=[1920,1080],display_size=[1920,1080],debug_mode=0):
super().__init__(kmodel_path,model_input_size,rgb888p_size,debug_mode)
# kmodel路径
self.kmodel_path=kmodel_path
# 检测模型输入分辨率
self.model_input_size=model_input_size
# sensor给到AI的图像分辨率，宽16字节对齐
self.rgb888p_size=[ALIGN_UP(rgb888p_size[0],16),rgb888p_size[1]]
# 视频输出VO分辨率，宽16字节对齐
self.display_size=[ALIGN_UP(display_size[0],16),display_size[1]]
self.crop_params=[]
# debug模式
self.debug_mode=debug_mode
# Ai2d实例用于实现预处理
self.ai2d=Ai2d(debug_mode)
# 设置ai2d的输入输出的格式和数据类型
self.ai2d.set_ai2d_dtype(nn.ai2d_format.NCHW_FMT,nn.ai2d_format.NCHW_FMT,np.uint8, np.uint8)
# 配置预处理操作，这里使用了crop和resize，Ai2d支持crop/shift/pad/resize/affine，具体代码请打开/sdcard/app/libs/AI2D.py查看
def config_preprocess(self,det,input_image_size=None):
with ScopedTiming("set preprocess config",self.debug_mode > 0):
ai2d_input_size=input_image_size if input_image_size else self.rgb888p_size
self.crop_params = self.get_crop_param(det)
self.ai2d.crop(self.crop_params[0],self.crop_params[1],self.crop_params[2],self.crop_params[3])
self.ai2d.resize(nn.interp_method.tf_bilinear, nn.interp_mode.half_pixel)
self.ai2d.build([1,3,ai2d_input_size[1],ai2d_input_size[0]],[1,3,self.model_input_size[1],self.model_input_size[0]])
# 自定义后处理，得到手掌手势结果和手掌关键点数据
def postprocess(self,results):
with ScopedTiming("postprocess",self.debug_mode > 0):
results=results[0].reshape(results[0].shape[0]*results[0].shape[1])
results_show = np.zeros(results.shape,dtype=np.int16)
results_show[0::2] = results[0::2] * self.crop_params[3] + self.crop_params[0]
results_show[1::2] = results[1::2] * self.crop_params[2] + self.crop_params[1]
gesture=self.hk_gesture(results_show)
results_show[0::2] = results_show[0::2] * (self.display_size[0] / self.rgb888p_size[0])
results_show[1::2] = results_show[1::2] * (self.display_size[1] / self.rgb888p_size[1])
return results_show,gesture
# 计算crop参数
def get_crop_param(self,det_box):
x1, y1, x2, y2 = det_box[2],det_box[3],det_box[4],det_box[5]
w,h= int(x2 - x1),int(y2 - y1)
w_det = int(float(x2 - x1) * self.display_size[0] // self.rgb888p_size[0])
h_det = int(float(y2 - y1) * self.display_size[1] // self.rgb888p_size[1])
x_det = int(x1*self.display_size[0] // self.rgb888p_size[0])
y_det = int(y1*self.display_size[1] // self.rgb888p_size[1])
length = max(w, h)/2
cx = (x1+x2)/2
cy = (y1+y2)/2
ratio_num = 1.26*length
x1_kp = int(max(0,cx-ratio_num))
y1_kp = int(max(0,cy-ratio_num))
x2_kp = int(min(self.rgb888p_size[0]-1, cx+ratio_num))
y2_kp = int(min(self.rgb888p_size[1]-1, cy+ratio_num))
w_kp = int(x2_kp - x1_kp + 1)
h_kp = int(y2_kp - y1_kp + 1)
return [x1_kp, y1_kp, w_kp, h_kp]
# 求两个vector之间的夹角
def hk_vector_2d_angle(self,v1,v2):
with ScopedTiming("hk_vector_2d_angle",self.debug_mode > 0):
v1_x,v1_y,v2_x,v2_y = v1[0],v1[1],v2[0],v2[1]
v1_norm = np.sqrt(v1_x * v1_x+ v1_y * v1_y)
v2_norm = np.sqrt(v2_x * v2_x + v2_y * v2_y)
dot_product = v1_x * v2_x + v1_y * v2_y
cos_angle = dot_product/(v1_norm*v2_norm)
angle = np.acos(cos_angle)*180/np.pi
return angle
# 根据手掌关键点检测结果判断手势类别
def hk_gesture(self,results):
with ScopedTiming("hk_gesture",self.debug_mode > 0):
angle_list = []
for i in range(5):
angle = self.hk_vector_2d_angle([(results[0]-results[i*8+4]), (results[1]-results[i*8+5])],[(results[i*8+6]-results[i*8+8]),(results[i*8+7]-results[i*8+9])])
angle_list.append(angle)
thr_angle,thr_angle_thumb,thr_angle_s,gesture_str = 65.,53.,49.,None
if 65535. not in angle_list:
if (angle_list[0]>thr_angle_thumb) and (angle_list[1]>thr_angle) and (angle_list[2]>thr_angle) and (angle_list[3]>thr_angle) and (angle_list[4]>thr_angle):
gesture_str = "fist"
elif (angle_list[0]<thr_angle_s) and (angle_list[1]<thr_angle_s) and (angle_list[2]<thr_angle_s) and (angle_list[3]<thr_angle_s) and (angle_list[4]<thr_angle_s):
gesture_str = "five"
elif (angle_list[0]<thr_angle_s) and (angle_list[1]<thr_angle_s) and (angle_list[2]>thr_angle) and (angle_list[3]>thr_angle) and (angle_list[4]>thr_angle):
gesture_str = "gun"
elif (angle_list[0]<thr_angle_s) and (angle_list[1]<thr_angle_s) and (angle_list[2]>thr_angle) and (angle_list[3]>thr_angle) and (angle_list[4]<thr_angle_s):
gesture_str = "love"
elif (angle_list[0]>5) and (angle_list[1]<thr_angle_s) and (angle_list[2]>thr_angle) and (angle_list[3]>thr_angle) and (angle_list[4]>thr_angle):
gesture_str = "one"
elif (angle_list[0]<thr_angle_s) and (angle_list[1]>thr_angle) and (angle_list[2]>thr_angle) and (angle_list[3]>thr_angle) and (angle_list[4]<thr_angle_s):
gesture_str = "six"
elif (angle_list[0]>thr_angle_thumb) and (angle_list[1]<thr_angle_s) and (angle_list[2]<thr_angle_s) and (angle_list[3]<thr_angle_s) and (angle_list[4]>thr_angle):
gesture_str = "three"
elif (angle_list[0]<thr_angle_s) and (angle_list[1]>thr_angle) and (angle_list[2]>thr_angle) and (angle_list[3]>thr_angle) and (angle_list[4]>thr_angle):
gesture_str = "thumbUp"
elif (angle_list[0]>thr_angle_thumb) and (angle_list[1]<thr_angle_s) and (angle_list[2]<thr_angle_s) and (angle_list[3]>thr_angle) and (angle_list[4]>thr_angle):
gesture_str = "yeah"
return gesture_str
# 手掌关键点分类任务
class HandKeyPointClass:
def __init__(self,hand_det_kmodel,hand_kp_kmodel,det_input_size,kp_input_size,labels,anchors,confidence_threshold=0.25,nms_threshold=0.3,nms_option=False,strides=[8,16,32],rgb888p_size=[1280,720],display_size=[1920,1080],debug_mode=0):
# 手掌检测模型路径
self.hand_det_kmodel=hand_det_kmodel
# 手掌关键点模型路径
self.hand_kp_kmodel=hand_kp_kmodel
# 手掌检测模型输入分辨率
self.det_input_size=det_input_size
# 手掌关键点模型输入分辨率
self.kp_input_size=kp_input_size
self.labels=labels
# anchors
self.anchors=anchors
# 置信度阈值
self.confidence_threshold=confidence_threshold
# nms阈值
self.nms_threshold=nms_threshold
self.nms_option=nms_option
self.strides=strides
# sensor给到AI的图像分辨率，宽16字节对齐
self.rgb888p_size=[ALIGN_UP(rgb888p_size[0],16),rgb888p_size[1]]
# 视频输出VO分辨率，宽16字节对齐
self.display_size=[ALIGN_UP(display_size[0],16),display_size[1]]
# debug_mode模式
self.debug_mode=debug_mode
self.hand_det=HandDetApp(self.hand_det_kmodel,self.labels,model_input_size=self.det_input_size,anchors=self.anchors,confidence_threshold=self.confidence_threshold,nms_threshold=self.nms_threshold,nms_option=self.nms_option,strides=self.strides,rgb888p_size=self.rgb888p_size,display_size=self.display_size,debug_mode=0)
self.hand_kp=HandKPClassApp(self.hand_kp_kmodel,model_input_size=self.kp_input_size,rgb888p_size=self.rgb888p_size,display_size=self.display_size)
self.hand_det.config_preprocess()
# run函数
def run(self,input_np):
# 执行手掌检测
det_boxes=self.hand_det.run(input_np)
boxes=[]
gesture_res=[]
for det_box in det_boxes:
# 对于检测到的每一个手掌执行关键点识别
x1, y1, x2, y2 = det_box[2],det_box[3],det_box[4],det_box[5]
w,h= int(x2 - x1),int(y2 - y1)
if (h<(0.1*self.rgb888p_size[1])):
continue
if (w<(0.25*self.rgb888p_size[0]) and ((x1<(0.03*self.rgb888p_size[0])) or (x2>(0.97*self.rgb888p_size[0])))):
continue
if (w<(0.15*self.rgb888p_size[0]) and ((x1<(0.01*self.rgb888p_size[0])) or (x2>(0.99*self.rgb888p_size[0])))):
continue
self.hand_kp.config_preprocess(det_box)
results_show,gesture=self.hand_kp.run(input_np)
gesture_res.append((results_show,gesture))
boxes.append(det_box)
return boxes,gesture_res
# 绘制效果，绘制关键点、手掌检测框和识别结果
def draw_result(self,pl,dets,gesture_res):
pl.osd_img.clear()
if len(dets)>0:
for k in range(len(dets)):
det_box=dets[k]
x1, y1, x2, y2 = det_box[2],det_box[3],det_box[4],det_box[5]
w,h= int(x2 - x1),int(y2 - y1)
if (h<(0.1*self.rgb888p_size[1])):
continue
if (w<(0.25*self.rgb888p_size[0]) and ((x1<(0.03*self.rgb888p_size[0])) or (x2>(0.97*self.rgb888p_size[0])))):
continue
if (w<(0.15*self.rgb888p_size[0]) and ((x1<(0.01*self.rgb888p_size[0])) or (x2>(0.99*self.rgb888p_size[0])))):
continue
w_det = int(float(x2 - x1) * self.display_size[0] // self.rgb888p_size[0])
h_det = int(float(y2 - y1) * self.display_size[1] // self.rgb888p_size[1])
x_det = int(x1*self.display_size[0] // self.rgb888p_size[0])
y_det = int(y1*self.display_size[1] // self.rgb888p_size[1])
pl.osd_img.draw_rectangle(x_det, y_det, w_det, h_det, color=(255, 0, 255, 0), thickness = 2)
results_show=gesture_res[k][0]
for i in range(len(results_show)/2):
pl.osd_img.draw_circle(results_show[i*2], results_show[i*2+1], 1, color=(255, 0, 255, 0),fill=False)
for i in range(5):
j = i*8
if i==0:
R = 255; G = 0; B = 0
if i==1:
R = 255; G = 0; B = 255
if i==2:
R = 255; G = 255; B = 0
if i==3:
R = 0; G = 255; B = 0
if i==4:
R = 0; G = 0; B = 255
pl.osd_img.draw_line(results_show[0], results_show[1], results_show[j+2], results_show[j+3], color=(255,R,G,B), thickness = 3)
pl.osd_img.draw_line(results_show[j+2], results_show[j+3], results_show[j+4], results_show[j+5], color=(255,R,G,B), thickness = 3)
pl.osd_img.draw_line(results_show[j+4], results_show[j+5], results_show[j+6], results_show[j+7], color=(255,R,G,B), thickness = 3)
pl.osd_img.draw_line(results_show[j+6], results_show[j+7], results_show[j+8], results_show[j+9], color=(255,R,G,B), thickness = 3)
gesture_str=gesture_res[k][1]
pl.osd_img.draw_string_advanced( x_det , y_det-50,32, " " + str(gesture_str), color=(255,0, 255, 0))
def audio_output_thread(output_stream):
CHUNK = int(0.3 * 16000) # 每次读取音频数据的帧数，设置为0.3s的帧数16000*0.3=4800
reply_wav_file = "/sdcard/examples/test.wav" # 音频路径
while audio_run:
if latest_gesture == "fist":
wf = wave.open(reply_wav_file, "rb")
wav_data = wf.read_frames(CHUNK)
while wav_data:
output_stream.write(wav_data)
wav_data = wf.read_frames(CHUNK)
time.sleep(1) # 时间缓冲，用于播放回复声音
wf.close()
if __name__ == "__main__":
# 其它参数
THRESH = 0.5 # 检测阈值
SAMPLE_RATE = 16000 # 采样率16000Hz,即每秒采样16000次
CHANNELS = 1 # 通道数 1为单声道，2为立体声
FORMAT = paInt16 # 音频输入输出格式 paInt16
CHUNK = int(0.3 * 16000) # 每次读取音频数据的帧数，设置为0.3s的帧数16000*0.3=4800
# 初始化音频流
p = PyAudio()
p.initialize(CHUNK)
# 显示模式，默认"hdmi",可以选择"hdmi"和"lcd"
display_mode = "lcd"
if display_mode == "hdmi":
display_size = [1920, 1080]
else:
display_size = [800, 480]
# 手掌检测模型路径
hand_det_kmodel_path = "/sdcard/examples/kmodel/hand_det.kmodel"
# 手掌关键点模型路径
hand_kp_kmodel_path = "/sdcard/examples/kmodel/handkp_det.kmodel"
# 其他参数
anchors_path = "/sdcard/examples/utils/prior_data_320.bin"
rgb888p_size = [1920, 1080]
hand_det_input_size = [512, 512]
hand_kp_input_size = [256, 256]
confidence_threshold = 0.2
nms_threshold = 0.5
labels = ["hand"]
anchors = [26, 27, 53, 52, 75, 71, 80, 99, 106, 82, 99, 134, 140, 113, 161, 172, 245, 276]
# 初始化PipeLine，只关注传给AI的图像分辨率，显示的分辨率
pl = PipeLine(rgb888p_size=rgb888p_size, display_size=display_size, display_mode=display_mode)
pl.create()
hkc = HandKeyPointClass(hand_det_kmodel_path, hand_kp_kmodel_path, det_input_size=hand_det_input_size, kp_input_size=hand_kp_input_size, labels=labels, anchors=anchors, confidence_threshold=confidence_threshold, nms_threshold=nms_threshold, nms_option=False, strides=[8, 16, 32], rgb888p_size=rgb888p_size, display_size=display_size, debug_mode=0)
clock = time.clock()
output_stream = p.open(format=FORMAT,channels=CHANNELS,rate=SAMPLE_RATE,output=True,frames_per_buffer=CHUNK)
_thread.start_new_thread(audio_output_thread,(output_stream,))
try:
while True:
os.exitpoint()
clock.tick()
img = pl.get_frame() # 获取当前帧
det_boxes, gesture_res = hkc.run(img) # 推理当前帧
hkc.draw_result(pl, det_boxes, gesture_res) # 绘制当前帧推理结果
print(det_boxes, gesture_res)
pl.show_image() # 展示推理结果
gc.collect()
print(clock.fps()) # 打印帧率
except Exception as e:
sys.print_exception(e)
audio_run=False
finally:
audio_run=False
time.sleep(1)
hkc.hand_det.deinit()
hkc.hand_kp.deinit()
pl.destroy()
以上是我的源代码
'''
from libs.PipeLine import PipeLine, ScopedTiming
from libs.AIBase import AIBase
from libs.AI2D import Ai2d
import os
import ujson
from media.media import * # 导入media模块，用于初始化vb buffer
from media.pyaudio import * # 导入pyaudio模块，用于采集和播放音频
import media.wave as wave # 导入wav模块，用于保存和加载wav音频文件
import nncase_runtime as nn
import ulab.numpy as np
import time
import image
import aicube
import random
import gc
import sys
def exit_check():
try:
os.exitpoint()
except KeyboardInterrupt as e:
print("user stop: ", e)
return True
return False
def play_audio(filename):
try:
wf = wave.open(filename, 'rb') # 打开wav文件
CHUNK = int(wf.get_framerate() / 25) # 设置音频chunk值
p = PyAudio()
p.initialize(CHUNK) # 初始化PyAudio对象
MediaManager.init() # vb buffer初始化
# 创建音频输出流，设置的音频参数均为wave中获取到的参数
stream = p.open(format=p.get_format_from_width(wf.get_sampwidth()),
channels=wf.get_channels(),
rate=wf.get_framerate(),
output=True, frames_per_buffer=CHUNK)
# 设置音频输出流的音量
stream.volume(vol=85)
data = wf.read_frames(CHUNK) # 从wav文件中读取数一帧数据
while data:
stream.write(data) # 将帧数据写入到音频输出流中
data = wf.read_frames(CHUNK) # 从wav文件中读取数一帧数据
if exit_check():
break
except BaseException as e:
print(f"Exception {e}")
finally:
stream.stop_stream() # 停止音频输出流
stream.close() # 关闭音频输出流
p.terminate() # 释放音频对象
wf.close() # 关闭wav文件
MediaManager.deinit() # 释放vb buffer
if __name__ == "__main__":
print("音频输出示例开始")
play_audio('/sdcard/examples/test.wav') # 播放WAV文件
print("音频输出示例完成")
'''
好的，太感谢了，在这个代码里是可以播放的test.wav文件