【CanMV K230 AI视觉】人脸关键部位

- 人脸关键部位

（动态测试效果可以去下面网站自己看。）

B站视频链接：已做成合集
抖音链接：已做成合集

人脸关键部位

人脸关键部位检测，主要检测脸部轮廓、眉毛、眼睛、鼻子和嘴巴，识别后通过画图示意出来，支持单个和多个人脸
在这里插入图片描述

'''
实验名称：人脸关键部位
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 *
from time import *
import nncase_runtime as nn
import ulab.numpy as np
import time
import image
import aidemo
import random
import gc
import sys# 自定义人脸检测任务类
class FaceDetApp(AIBase):def __init__(self,kmodel_path,model_input_size,anchors,confidence_threshold=0.25,nms_threshold=0.3,rgb888p_size=[1280,720],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# 置信度阈值self.confidence_threshold=confidence_threshold# nms阈值self.nms_threshold=nms_threshold# 检测任务锚框self.anchors=anchors# 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# 实例化Ai2d，用于实现模型预处理self.ai2d=Ai2d(debug_mode)# 设置Ai2d的输入输出格式和类型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# 设置padding预处理self.ai2d.pad(self.get_pad_param(), 0, [104,117,123])# 设置resize预处理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shapeself.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，results是模型输出的array列表，这里使用了aidemo的face_det_post_process列表def postprocess(self,results):with ScopedTiming("postprocess",self.debug_mode > 0):res = aidemo.face_det_post_process(self.confidence_threshold,self.nms_threshold,self.model_input_size[0],self.anchors,self.rgb888p_size,results)if len(res)==0:return reselse:return res[0]# 计算padding参数def get_pad_param(self):dst_w = self.model_input_size[0]dst_h = self.model_input_size[1]# 计算最小的缩放比例，等比例缩放ratio_w = dst_w / self.rgb888p_size[0]ratio_h = dst_h / self.rgb888p_size[1]if ratio_w < ratio_h:ratio = ratio_welse:ratio = ratio_hnew_w = (int)(ratio * self.rgb888p_size[0])new_h = (int)(ratio * self.rgb888p_size[1])dw = (dst_w - new_w) / 2dh = (dst_h - new_h) / 2top = (int)(round(0))bottom = (int)(round(dh * 2 + 0.1))left = (int)(round(0))right = (int)(round(dw * 2 - 0.1))return [0,0,0,0,top, bottom, left, right]# 自定义人脸关键点任务类
class FaceLandMarkApp(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# 关键点模型输入分辨率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]]# debug模式self.debug_mode=debug_mode# 目标矩阵self.matrix_dst=Noneself.ai2d=Ai2d(debug_mode)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affine，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):# 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# 计算目标矩阵，并获取仿射变换矩阵self.matrix_dst = self.get_affine_matrix(det)affine_matrix = [self.matrix_dst[0][0],self.matrix_dst[0][1],self.matrix_dst[0][2],self.matrix_dst[1][0],self.matrix_dst[1][1],self.matrix_dst[1][2]]# 设置仿射变换预处理self.ai2d.affine(nn.interp_method.cv2_bilinear,0, 0, 127, 1,affine_matrix)# 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shapeself.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，results是模型输出的array列表，这里使用了aidemo库的invert_affine_transform接口def postprocess(self,results):with ScopedTiming("postprocess",self.debug_mode > 0):pred=results[0]# （1）将人脸关键点输出变换模型输入half_input_len = self.model_input_size[0] // 2pred = pred.flatten()for i in range(len(pred)):pred[i] += (pred[i] + 1) * half_input_len# （2）获取仿射矩阵的逆矩阵matrix_dst_inv = aidemo.invert_affine_transform(self.matrix_dst)matrix_dst_inv = matrix_dst_inv.flatten()# （3）对每个关键点进行逆变换half_out_len = len(pred) // 2for kp_id in range(half_out_len):old_x = pred[kp_id * 2]old_y = pred[kp_id * 2 + 1]# 逆变换公式new_x = old_x * matrix_dst_inv[0] + old_y * matrix_dst_inv[1] + matrix_dst_inv[2]new_y = old_x * matrix_dst_inv[3] + old_y * matrix_dst_inv[4] + matrix_dst_inv[5]pred[kp_id * 2] = new_xpred[kp_id * 2 + 1] = new_yreturn preddef get_affine_matrix(self,bbox):# 获取仿射矩阵，用于将边界框映射到模型输入空间with ScopedTiming("get_affine_matrix", self.debug_mode > 1):# 从边界框提取坐标和尺寸x1, y1, w, h = map(lambda x: int(round(x, 0)), bbox[:4])# 计算缩放比例，使得边界框映射到模型输入空间的一部分scale_ratio = (self.model_input_size[0]) / (max(w, h) * 1.5)# 计算边界框中心点在模型输入空间的坐标cx = (x1 + w / 2) * scale_ratiocy = (y1 + h / 2) * scale_ratio# 计算模型输入空间的一半长度half_input_len = self.model_input_size[0] / 2# 创建仿射矩阵并进行设置matrix_dst = np.zeros((2, 3), dtype=np.float)matrix_dst[0, 0] = scale_ratiomatrix_dst[0, 1] = 0matrix_dst[0, 2] = half_input_len - cxmatrix_dst[1, 0] = 0matrix_dst[1, 1] = scale_ratiomatrix_dst[1, 2] = half_input_len - cyreturn matrix_dst# 人脸标志解析
class FaceLandMark:def __init__(self,face_det_kmodel,face_landmark_kmodel,det_input_size,landmark_input_size,anchors,confidence_threshold=0.25,nms_threshold=0.3,rgb888p_size=[1920,1080],display_size=[1920,1080],debug_mode=0):# 人脸检测模型路径self.face_det_kmodel=face_det_kmodel# 人脸标志解析模型路径self.face_landmark_kmodel=face_landmark_kmodel# 人脸检测模型输入分辨率self.det_input_size=det_input_size# 人脸标志解析模型输入分辨率self.landmark_input_size=landmark_input_size# anchorsself.anchors=anchors# 置信度阈值self.confidence_threshold=confidence_threshold# nms阈值self.nms_threshold=nms_threshold# 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# 人脸关键点不同部位关键点列表self.dict_kp_seq = [[43, 44, 45, 47, 46, 50, 51, 49, 48],              # left_eyebrow[97, 98, 99, 100, 101, 105, 104, 103, 102],        # right_eyebrow[35, 36, 33, 37, 39, 42, 40, 41],                  # left_eye[89, 90, 87, 91, 93, 96, 94, 95],                  # right_eye[34, 88],                                          # pupil[72, 73, 74, 86],                                  # bridge_nose[77, 78, 79, 80, 85, 84, 83],                      # wing_nose[52, 55, 56, 53, 59, 58, 61, 68, 67, 71, 63, 64],  # out_lip[65, 54, 60, 57, 69, 70, 62, 66],                  # in_lip[1, 9, 10, 11, 12, 13, 14, 15, 16, 2, 3, 4, 5, 6, 7, 8, 0, 24, 23, 22, 21, 20, 19, 18, 32, 31, 30, 29, 28, 27, 26, 25, 17]  # basin]# 人脸关键点不同部位（顺序同dict_kp_seq）颜色配置，argbself.color_list_for_osd_kp = [(255, 0, 255, 0),(255, 0, 255, 0),(255, 255, 0, 255),(255, 255, 0, 255),(255, 255, 0, 0),(255, 255, 170, 0),(255, 255, 255, 0),(255, 0, 255, 255),(255, 255, 220, 50),(255, 30, 30, 255)]# 人脸检测实例self.face_det=FaceDetApp(self.face_det_kmodel,model_input_size=self.det_input_size,anchors=self.anchors,confidence_threshold=self.confidence_threshold,nms_threshold=self.nms_threshold,rgb888p_size=self.rgb888p_size,display_size=self.display_size,debug_mode=0)# 人脸标志解析实例self.face_landmark=FaceLandMarkApp(self.face_landmark_kmodel,model_input_size=self.landmark_input_size,rgb888p_size=self.rgb888p_size,display_size=self.display_size)# 配置人脸检测的预处理self.face_det.config_preprocess()# run函数def run(self,input_np):# 执行人脸检测det_boxes=self.face_det.run(input_np)landmark_res=[]for det_box in det_boxes:# 对每一个检测到的人脸解析关键部位self.face_landmark.config_preprocess(det_box)res=self.face_landmark.run(input_np)landmark_res.append(res)return det_boxes,landmark_res# 绘制人脸解析效果def draw_result(self,pl,dets,landmark_res):pl.osd_img.clear()if dets:draw_img_np = np.zeros((self.display_size[1],self.display_size[0],4),dtype=np.uint8)draw_img = image.Image(self.display_size[0], self.display_size[1], image.ARGB8888, alloc=image.ALLOC_REF,data = draw_img_np)for pred in landmark_res:# （1）获取单个人脸框对应的人脸关键点for sub_part_index in range(len(self.dict_kp_seq)):# （2）构建人脸某个区域关键点集sub_part = self.dict_kp_seq[sub_part_index]face_sub_part_point_set = []for kp_index in range(len(sub_part)):real_kp_index = sub_part[kp_index]x, y = pred[real_kp_index * 2], pred[real_kp_index * 2 + 1]x = int(x * self.display_size[0] // self.rgb888p_size[0])y = int(y * self.display_size[1] // self.rgb888p_size[1])face_sub_part_point_set.append((x, y))# （3）画人脸不同区域的轮廓if sub_part_index in (9, 6):color = np.array(self.color_list_for_osd_kp[sub_part_index],dtype = np.uint8)face_sub_part_point_set = np.array(face_sub_part_point_set)aidemo.polylines(draw_img_np, face_sub_part_point_set,False,color,5,8,0)elif sub_part_index == 4:color = self.color_list_for_osd_kp[sub_part_index]for kp in face_sub_part_point_set:x,y = kp[0],kp[1]draw_img.draw_circle(x,y ,2, color, 1)else:color = np.array(self.color_list_for_osd_kp[sub_part_index],dtype = np.uint8)face_sub_part_point_set = np.array(face_sub_part_point_set)aidemo.contours(draw_img_np, face_sub_part_point_set,-1,color,2,8)pl.osd_img.copy_from(draw_img)if __name__=="__main__":# 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]# 人脸检测模型路径face_det_kmodel_path="/sdcard/app/tests/kmodel/face_detection_320.kmodel"# 人脸关键标志模型路径face_landmark_kmodel_path="/sdcard/app/tests/kmodel/face_landmark.kmodel"# 其它参数anchors_path="/sdcard/app/tests/utils/prior_data_320.bin"rgb888p_size=[1920,1080]face_det_input_size=[320,320]face_landmark_input_size=[192,192]confidence_threshold=0.5nms_threshold=0.2anchor_len=4200det_dim=4anchors = np.fromfile(anchors_path, dtype=np.float)anchors = anchors.reshape((anchor_len,det_dim))# 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()flm=FaceLandMark(face_det_kmodel_path,face_landmark_kmodel_path,det_input_size=face_det_input_size,landmark_input_size=face_landmark_input_size,anchors=anchors,confidence_threshold=confidence_threshold,nms_threshold=nms_threshold,rgb888p_size=rgb888p_size,display_size=display_size)clock = time.clock()try:while True:os.exitpoint()clock.tick()img=pl.get_frame()                          # 获取当前帧det_boxes,landmark_res=flm.run(img)         # 推理当前帧print(det_boxes,landmark_res)  # 打印结果flm.draw_result(pl,det_boxes,landmark_res)  # 绘制推理结果pl.show_image()                             # 展示推理效果gc.collect()print(clock.fps()) #打印帧率except Exception as e:sys.print_exception(e)finally:flm.face_det.deinit()flm.face_landmark.deinit()pl.destroy()