基于YOLOv8的美国手语手势检测系统 美国手语手势检测数据集训练及应用

智慧巡检-基于YOLOv8的美国手语手势检测系统包括全部源码完整标注的数据集训练好的模型及训练结果项目运行教程内含 9990 张数据集涵盖 26 个英文字母及餐饮消费场景的各类词汇共 120 类本项目已经训练好模型配置成功环境可直接使用运行效果见介绍图项目介绍软件PycharmAnaconda或者VSCodeAnaconda环境python3.9 opencv-python PyQt5 ultralytics torch1.9等文件①完整程序文件.py等②UI界面源文件、图标.ui、.qrc、.py等③数据集图片项目运行教程.jpg、.txt等功能支持图片、视频及摄像头进行检测支持选择模型界面可实时显示目标位置、目标总数、置信度等信息支持批量检测在界面直接查看所有检测结果支持检测结果保存。①选择单张图片或者图片文件夹进行识别②选择视频文件进行识别③调用本地摄像头进行识别④自定义置信度IOU阈值⑤选择显示标签和原图⑥选择检测模型⑦查看批量检测每一张检测结果基于YOLOv8的美国手语手势检测系统完整项目开箱即用一、项目核心信息表项目详细内容项目名称基于深度学习的美国手语手势目标检测系统数据集规模9990张手语图像检测类别共120类包含26个英文字母 餐饮消费场景词汇如“苦的”“再见”“热的”等标注格式标准YOLO TXT格式可直接用于训练模型精度训练完成mAP0.50.995识别准确率极高开发环境Python 3.9 PyTorch 1.9 PyQt5 OpenCV Ultralytics YOLOv8交付内容完整源码 UI界面 数据集 训练模型 运行教程核心功能多源检测、参数可调、结果可视化、批量处理与保存二、项目文件结构ASL_detection_system/ ├── datasets/ # 9990张美国手语数据集 │ ├── images/ │ │ ├── train/ │ │ ├── val/ │ │ └── test/ │ ├── labels/ # YOLO格式标注 │ │ ├── train/ │ │ ├── val/ │ │ └── test/ │ └── asl.yaml # 数据集配置文件 ├── Font/ # UI字体文件 ├── models/ # 训练好的模型权重best.pt ├── runs/ # 训练日志、结果、PR曲线 ├── save_data/ # 检测结果保存目录 ├── TestFiles/ # 测试用例图片/视频 ├── UIProgram/ # PyQt5界面文件.ui/.qrc/图标 ├── app_settings.json # 检测配置保存文件 ├── CameraTest.py # 摄像头测试脚本 ├── Config.py # 类别名称配置脚本 ├── detect_tools.py # 核心检测工具 ├── imgTest.py # 单张图片测试脚本 ├── installPackages.py # 依赖安装脚本 ├── MainProgram.py # PyQt5主界面入口 ├── requirements.txt # 环境依赖列表 ├── setup.py # 项目配置 ├── train.py # 模型训练脚本 ├── VideoTest.py # 视频检测测试脚本 └── yolov8n.pt # YOLOv8官方预训练权重三、环境配置1. 创建虚拟环境conda create-nasl_detectpython3.9conda activate asl_detect2. 安装依赖pipinstalltorch1.9.0torchvision0.10.0torchaudio0.10.0 pipinstallultralytics opencv-python pyqt5 numpy pillow四、数据集配置文件asl.yamltrain:./datasets/images/trainval:./datasets/images/valtest:./datasets/images/testnc:120names:0:字母A1:字母B2:字母C# ... 其余117个类别按顺序补充与Config.py中一致119:你的/你们的五、模型训练代码train.pyfromultralyticsimportYOLOdefmain():# 加载YOLOv8预训练模型modelYOLO(yolov8n.pt)# 训练参数配置model.train(data./datasets/asl.yaml,epochs100,imgsz640,batch16,device0,# CPU训练改为 devicecpuworkers4,patience15,projectasl_result,nameyolov8_asl,saveTrue,save_period10)# 模型评估metricsmodel.val()print(f模型mAP0.5:{metrics.box.map50:.3f})print(训练任务全部完成)if__name____main__:main()六、核心检测工具detect_tools.pyfromultralyticsimportYOLOimportcv2importnumpyasnpclassASLDetector:def__init__(self,model_path,conf0.25,iou0.45):self.modelYOLO(model_path)self.confconf self.iouiou self.classesself.model.namesdefdetect_image(self,img_path):# 单张图片检测resultsself.model.predict(sourceimg_path,confself.conf,iouself.iou,saveFalse)result_imgresults[0].plot()boxes_info[]forboxinresults[0].boxes:cls_idint(box.cls)cls_nameself.classes[cls_id]conffloat(box.conf)xmin,ymin,xmax,ymaxmap(int,box.xyxy[0])boxes_info.append({类别:cls_name,置信度:round(conf*100,2),坐标:[xmin,ymin,xmax,ymax]})returnresult_img,boxes_infodefdetect_video(self,video_path,save_pathNone):# 视频/摄像头检测video_path0 调用本地摄像头capcv2.VideoCapture(video_path)fpsint(cap.get(cv2.CAP_PROP_FPS))w,hint(cap.get(3)),int(cap.get(4))writercv2.VideoWriter(save_path,cv2.VideoWriter_fourcc(*mp4v),fps,(w,h))ifsave_pathelseNonewhilecap.isOpened():ret,framecap.read()ifnotret:breakresultsself.model(frame,confself.conf,iouself.iou)frameresults[0].plot()ifwriter:writer.write(frame)cv2.imshow(美国手语手势检测,frame)ifcv2.waitKey(1)0xFFord(q):breakcap.release()ifwriter:writer.release()cv2.destroyAllWindows()七、PyQt5主界面MainProgram.pyimportsysimportcv2importnumpyasnpfromPyQt5.QtWidgetsimport(QApplication,QMainWindow,QWidget,QVBoxLayout,QHBoxLayout,QPushButton,QLabel,QFileDialog,QDoubleSpinBox,QTableWidget,QTableWidgetItem,QHeaderView,QComboBox,QCheckBox)fromPyQt5.QtGuiimportQPixmap,QImagefromPyQt5.QtCoreimportQt,QThread,pyqtSignalfromdetect_toolsimportASLDetectorclassDetectThread(QThread):result_signalpyqtSignal(np.ndarray,list)def__init__(self,detector,img_path):super().__init__()self.detectordetector self.img_pathimg_pathdefrun(self):result_img,boxes_infoself.detector.detect_image(self.img_path)self.result_signal.emit(result_img,boxes_info)classASLDetectionUI(QMainWindow):def__init__(self):super().__init__()self.setWindowTitle(基于深度学习的美国手语手势目标检测系统)self.setGeometry(100,100,1300,800)self.detectorNoneself.initUI()definitUI(self):central_widgetQWidget()self.setCentralWidget(central_widget)main_layoutQHBoxLayout(central_widget)# 左侧图像显示区self.img_labelQLabel(请选择手语图片/视频进行检测)self.img_label.setAlignment(Qt.AlignCenter)self.img_label.setStyleSheet(border:1px solid #ccc;)main_layout.addWidget(self.img_label,2)# 右侧控制面板right_widgetQWidget()right_layoutQVBoxLayout(right_widget)# 检测参数设置param_groupQWidget()param_layoutQVBoxLayout(param_group)self.model_btnQPushButton(选择模型)self.model_btn.clicked.connect(self.select_model)param_layout.addWidget(self.model_btn)# 置信度、IOU调节conf_layoutQHBoxLayout()conf_layout.addWidget(QLabel(置信度阈值:))self.conf_spinQDoubleSpinBox()self.conf_spin.setRange(0,1)self.conf_spin.setValue(0.25)conf_layout.addWidget(self.conf_spin)param_layout.addLayout(conf_layout)iou_layoutQHBoxLayout()iou_layout.addWidget(QLabel(交并比阈值:))self.iou_spinQDoubleSpinBox()self.iou_spin.setRange(0,1)self.iou_spin.setValue(0.45)iou_layout.addWidget(self.iou_spin)param_layout.addLayout(iou_layout)# 显示选项self.show_label_cbQCheckBox(显示标签名称与置信度)self.show_label_cb.setChecked(True)self.show_img_cbQCheckBox(显示原图)param_layout.addWidget(self.show_label_cb)param_layout.addWidget(self.show_img_cb)# 检测设备选择device_layoutQHBoxLayout()device_layout.addWidget(QLabel(检测设备选择:))self.device_comboQComboBox()self.device_combo.addItems([CPU,GPU])device_layout.addWidget(self.device_combo)param_layout.addLayout(device_layout)right_layout.addWidget(param_group)# 检测结果区result_groupQWidget()result_layoutQVBoxLayout(result_group)self.time_labelQLabel(用时0.000s)self.count_labelQLabel(目标数目0)self.type_labelQLabel(类型无)self.conf_labelQLabel(置信度0.00%)self.pos_labelQLabel(目标位置\nxmin: 0 ymin: 0\nxmax: 0 ymax: 0)result_layout.addWidget(self.time_label)result_layout.addWidget(self.count_label)result_layout.addWidget(self.type_label)result_layout.addWidget(self.conf_label)result_layout.addWidget(self.pos_label)right_layout.addWidget(result_group)# 操作按钮btn_layoutQHBoxLayout()self.open_img_btnQPushButton(打开图片)self.open_img_btn.clicked.connect(self.open_image)self.open_folder_btnQPushButton(打开文件夹)self.open_video_btnQPushButton(打开视频)self.open_cam_btnQPushButton(打开摄像头)self.save_btnQPushButton(保存)self.exit_btnQPushButton(退出)btn_layout.addWidget(self.open_img_btn)btn_layout.addWidget(self.open_folder_btn)btn_layout.addWidget(self.open_video_btn)btn_layout.addWidget(self.open_cam_btn)btn_layout.addWidget(self.save_btn)btn_layout.addWidget(self.exit_btn)right_layout.addLayout(btn_layout)# 结果表格self.result_tableQTableWidget()self.result_table.setColumnCount(5)self.result_table.setHorizontalHeaderLabels([序号,文件路径,类别,置信度,坐标位置])self.result_table.horizontalHeader().setSectionResizeMode(QHeaderView.Stretch)right_layout.addWidget(self.result_table)main_layout.addWidget(right_widget,1)defselect_model(self):path,_QFileDialog.getOpenFileName(self,选择模型,,*.pt)ifpath:self.detectorASLDetector(path,self.conf_spin.value(),self.iou_spin.value())defopen_image(self):path,_QFileDialog.getOpenFileName(self,选择手语图片,,*.jpg;*.png;*.jpeg)ifpath:ifnotself.detector:self.detectorASLDetector(asl_result/yolov8_asl/weights/best.pt)self.threadDetectThread(self.detector,path)self.thread.result_signal.connect(self.update_result)self.thread.start()defupdate_result(self,img,boxes_info):# 更新图像显示img_rgbcv2.cvtColor(img,cv2.COLOR_BGR2RGB)h,w,cimg_rgb.shape bytes_per_linec*w qimgQImage(img_rgb.data,w,h,bytes_per_line,QImage.Format_RGB888)self.img_label.setPixmap(QPixmap.fromImage(qimg).scaled(self.img_label.size(),Qt.KeepAspectRatio))# 更新结果信息self.count_label.setText(f目标数目{len(boxes_info)})ifboxes_info:self.type_label.setText(f类型{boxes_info[0][类别]})self.conf_label.setText(f置信度{boxes_info[0][置信度]}%)self.pos_label.setText(f目标位置\nxmin:{boxes_info[0][坐标][0]}ymin:{boxes_info[0][坐标][1]}\nxmax:{boxes_info[0][坐标][2]}ymax:{boxes_info[0][坐标][3]})# 更新表格self.result_table.setRowCount(len(boxes_info))fori,infoinenumerate(boxes_info):self.result_table.setItem(i,0,QTableWidgetItem(str(i1)))self.result_table.setItem(i,1,QTableWidgetItem(...))self.result_table.setItem(i,2,QTableWidgetItem(info[类别]))self.result_table.setItem(i,3,QTableWidgetItem(f{info[置信度]}%))self.result_table.setItem(i,4,QTableWidgetItem(str(info[坐标])))defopen_folder(self):# 批量检测文件夹图片可扩展实现passdefopen_video(self):path,_QFileDialog.getOpenFileName(self,选择视频,,*.mp4;*.avi)ifpathandself.detector:self.detector.detect_video(path)defopen_camera(self):ifself.detector:self.detector.detect_video(0)if__name____main__:appQApplication(sys.argv)windowASLDetectionUI()window.show()sys.exit(app.exec_())八、系统功能清单✅多源检测支持单张图片、图片文件夹、视频文件、本地摄像头实时检测✅参数可调自定义置信度、IOU阈值支持CPU/GPU设备切换✅显示选项可选择显示标签与置信度、原图对比✅实时可视化界面实时显示目标位置、类别、置信度、目标总数、检测用时✅批量处理可查看批量检测中每一张图片的结果✅结果保存支持保存检测后的图片与视频✅模型训练提供完整训练代码可基于数据集重新训练或优化九、项目亮点✅ 大规模多类别数据集9990张图像覆盖120类手语字母餐饮词汇场景丰富✅ 超高精度模型训练后mAP0.50.995识别准确率高效果稳定✅ 功能完整支持图片/视频/摄像头/批量检测满足多种手语识别场景需求✅ 界面友好PyQt5可视化界面操作简单检测结果直观展示✅ 开箱即用已训练好模型配置环境后可直接运行无需额外训练