基于深度学习mediapipe的智能坐姿检测系统源码+模型.zip

import os import pickle import cv2 from colorama import Fore import numpy as np import Core.pose as pose posImagePath = 'Data/Train/Image/Pos/' negImagePath = 'Data/Train/Image/Neg/' resultsPath = 'Data/Train/Processed/' posDatas = [] negDatas = [] def gamma_trans(img,gamma): gamma_table = [np.power(x/255.0,gamma)*255.0 for x in range(256)] gamma_table = np.round(np.array(gamma_table)).astype(np.uint8) return cv2.LUT(img,gamma_table) def processImage(img): # cv2.imwrite('Data/Processed' + str(len(posDatas) + len(negDatas)) + '.jpg', img) imgRGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) poseData = pose.getPosePoints(imgRGB) if poseData: poseData = [poseData[0]] + poseData[11:16] '''xyMin, xyMax = 100000,- 1 zMin, zMax = 100000,- 1 for pointData in poseData: for i, pointValue in enumerate(pointData): if i in (0, 1): xyMax = max(xyMax, pointValue) xyMin = min(xyMin, pointValue) else: zMax = max(zMax, pointValue) zMin = min(zMin, pointValue) for pointData in poseData: for i, pointValue in enumerate(pointData): if i in (0, 1): pointData[i] = (pointData[i] - xyMin) / (xyMax - xyMin) else: pointData[i] = (pointData[i] - zMin) / (zMax - zMin)''' '''minValues, maxValues = [100000] * 3, [-1] * 3 for pointData in poseData: for i, pointValue in enumerate(pointData): minValues[i] = min(minValues[i], pointValue) maxValues[i] = max(maxValues[i], pointValue) for pointData in poseData: for i, pointValue in enumerate(pointData): pointData[i] = (pointData[i] - minValues[i%3]) / (maxValues[i%3] - minValues[i%3])''' return poseData else: print('Error') return None def process_(img, dataList): data = processImage(img) if data: dataList.append(data) img_ = cv2.flip(img, 1) data = processImage(img_) if data: dataList.append(data) img_ = gamma_trans(img, 1.2) data = processImage(img_) if data: dataList.append(data) img_ = gamma_trans(img, 0.8) data = processImage(img_) if data: dataList.append(data) def process(): print(Fore.RED + '加载并处理训练数据中...') print(Fore.RESET, end='') for path in os.listdir(posImagePath): img = cv2.imread(os.path.join(posImagePath, path)) process_(img, posDatas) for path in os.listdir(negImagePath): img = cv2.imread(os.path.join(negImagePath, path)) process_(img, negDatas) print(Fore.CYAN + '加载并处理完毕') print(Fore.RESET) print(f'Pos data length: {len(posDatas)}') print(f'Pos data size: {len(posDatas)} * {len(posDatas[0])} * {len(posDatas[0][0])}') print('图片数量 * 关键点数量 * 关键点坐标维度') print() print(f'Neg data length: {len(negDatas)}') print(f'Neg data size: {len(negDatas)} * {len(negDatas[0])} * {len(negDatas[0][0])}') print('图片数量 * 关键点数量 * 关键点坐标维度') print(Fore.RESET, end='') pickle.dump(posDatas, open(os.path.join(resultsPath, 'posData.data'), 'wb')) pickle.dump(negDatas, open(os.path.join(resultsPath, 'negData.data'), 'wb'))