435 lines
15 KiB
JavaScript
435 lines
15 KiB
JavaScript
/**
|
||
* =============================================================================
|
||
* 动态版 - 手部姿态识别与模型管理脚本 (v3.0)
|
||
* 由人体姿态识别修改为手部姿态识别
|
||
* =============================================================================
|
||
*/
|
||
|
||
'use strict';
|
||
|
||
// --- 全局变量和常量 ---
|
||
const videoElement = document.getElementById('video');
|
||
const canvasElement = document.getElementById('canvas');
|
||
const canvasCtx = canvasElement.getContext('2d');
|
||
const statusElement = document.getElementById('status');
|
||
const resultElement = document.getElementById('result-text');
|
||
|
||
// UI元素
|
||
const poseClassesContainer = document.getElementById('pose-classes-container');
|
||
const addClassButton = document.getElementById('btn-add-class');
|
||
const predictButton = document.getElementById('btn-predict');
|
||
const exportButton = document.getElementById('btn-export');
|
||
const importButton = document.getElementById('btn-import');
|
||
const fileImporter = document.getElementById('file-importer');
|
||
|
||
let detector, classifier, animationFrameId;
|
||
let isPredicting = false;
|
||
|
||
const appState = {
|
||
classMap: {},
|
||
nextClassId: 0
|
||
};
|
||
|
||
// --- 主应用逻辑 ---
|
||
|
||
/**
|
||
* 初始化应用，加载模型并设置摄像头
|
||
*/
|
||
async function init() {
|
||
try {
|
||
classifier = knnClassifier.create();
|
||
|
||
// --- 修改点 1: 加载手部检测模型 ---
|
||
const model = handPoseDetection.SupportedModels.MediaPipeHands;
|
||
const detectorConfig = {
|
||
runtime: 'mediapipe', // 推荐使用 MediaPipe runtime 获得最佳性能
|
||
solutionPath: 'https://cdn.jsdelivr.net/npm/@mediapipe/hands' // MediaPipe solution files path
|
||
};
|
||
detector = await handPoseDetection.createDetector(model, detectorConfig);
|
||
|
||
await setupCamera();
|
||
setupEventListeners();
|
||
mainLoop();
|
||
|
||
statusElement.innerText = "手部模型和摄像头已就绪!";
|
||
enableControls();
|
||
addNewClass(); // 默认创建第一个类别
|
||
|
||
} catch (error) {
|
||
console.error("初始化失败:", error);
|
||
statusElement.innerText = "初始化失败，请检查摄像头权限或刷新。";
|
||
statusElement.style.backgroundColor = '#fce8e6';
|
||
statusElement.style.color = '#d93025';
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 设置和启动用户摄像头 (无需修改，与之前一致)
|
||
*/
|
||
async function setupCamera() {
|
||
const stream = await navigator.mediaDevices.getUserMedia({ video: true });
|
||
videoElement.srcObject = stream;
|
||
return new Promise((resolve) => {
|
||
videoElement.onloadedmetadata = () => {
|
||
videoElement.play();
|
||
// 确保 Canvas 与 Video 宽高一致，并且在这里不需要 Canvas 镜像
|
||
canvasElement.width = videoElement.videoWidth;
|
||
canvasElement.height = videoElement.videoHeight;
|
||
resolve();
|
||
};
|
||
});
|
||
}
|
||
|
||
/**
|
||
* 为所有交互式元素绑定事件监听器 (无需修改，与之前一致)
|
||
*/
|
||
function setupEventListeners() {
|
||
addClassButton.addEventListener('click', addNewClass);
|
||
predictButton.addEventListener('click', togglePrediction);
|
||
exportButton.addEventListener('click', exportModel);
|
||
importButton.addEventListener('click', () => fileImporter.click());
|
||
fileImporter.addEventListener('change', importModel);
|
||
}
|
||
|
||
// --- 动态类别管理 (无需修改，与之前一致) ---
|
||
|
||
/**
|
||
* 动态创建一个新类别的UI元素并添加到页面
|
||
* @param {number} cId - 类别的唯一ID
|
||
* @param {string} cName - 类别的名称
|
||
*/
|
||
function createClassUI(cId, cName) {
|
||
const poseClassDiv = document.createElement('div');
|
||
poseClassDiv.className = 'pose-class';
|
||
poseClassDiv.dataset.classId = cId;
|
||
|
||
poseClassDiv.innerHTML = `
|
||
<div class="class-info">
|
||
<input type="text" class="class-name-input" value="${cName}" data-class-id="${cId}">
|
||
<span class="sample-count">(0 样本)</span>
|
||
</div>
|
||
<div class="class-actions">
|
||
<button class="btn-sample" data-class-id="${cId}">采集样本</button>
|
||
<button class="btn-delete-class" title="删除类别" data-class-id="${cId}">×</button>
|
||
</div>
|
||
`;
|
||
|
||
poseClassesContainer.appendChild(poseClassDiv);
|
||
|
||
const nameInput = poseClassDiv.querySelector('.class-name-input');
|
||
nameInput.addEventListener('change', (e) => {
|
||
appState.classMap[cId] = e.target.value;
|
||
});
|
||
|
||
const sampleButton = poseClassDiv.querySelector('.btn-sample');
|
||
sampleButton.addEventListener('click', () => addExample(cId));
|
||
|
||
if (isPredicting) sampleButton.disabled = true;
|
||
|
||
const deleteButton = poseClassDiv.querySelector('.btn-delete-class');
|
||
deleteButton.addEventListener('click', () => deleteClass(cId));
|
||
}
|
||
|
||
/**
|
||
* 添加一个新的姿态类别
|
||
*/
|
||
function addNewClass() {
|
||
const classId = appState.nextClassId;
|
||
const className = `手势 ${classId + 1}`; // 改为“手势”
|
||
appState.classMap[classId] = className;
|
||
appState.nextClassId++;
|
||
createClassUI(classId, className);
|
||
}
|
||
|
||
/**
|
||
* 删除一个指定的姿态类别
|
||
* @param {number} classId - 要删除的类别的ID
|
||
*/
|
||
function deleteClass(classId) {
|
||
if (confirm(`确定要删除类别 "${appState.classMap[classId]}" 吗？所有样本都将丢失。`)) {
|
||
const elementToRemove = poseClassesContainer.querySelector(`[data-class-id="${classId}"]`);
|
||
if (elementToRemove) elementToRemove.remove();
|
||
|
||
delete appState.classMap[classId];
|
||
classifier.clearClass(classId);
|
||
|
||
updateSampleCounts();
|
||
updatePredictionUI();
|
||
checkExportAbility();
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 采集一个姿态样本并添加到KNN分类器
|
||
* @param {number} classId 类别的ID
|
||
*/
|
||
async function addExample(classId) {
|
||
// --- 修改点 2: 使用 estimateHands 替代 estimatePoses ---
|
||
// flipHorizontal: false 确保模型输出的坐标与原视频方向一致 (非镜像)
|
||
const hands = await detector.estimateHands(videoElement, { flipHorizontal: false });
|
||
if (hands && hands.length > 0) {
|
||
// KNN 分类器通常只处理一个实例，这里我们取检测到的第一只手
|
||
const handTensor = flattenHand(hands[0]); // 使用新的 flattenHand
|
||
classifier.addExample(handTensor, classId);
|
||
handTensor.dispose(); // 释放内存
|
||
|
||
updateSampleCounts();
|
||
checkExportAbility();
|
||
} else {
|
||
console.warn(`为类别 ${appState.classMap[classId]} 采集样本失败，未检测到手部。`);
|
||
}
|
||
}
|
||
|
||
// --- 模型与预测逻辑 ---
|
||
|
||
/**
|
||
* 开始或停止姿态预测
|
||
*/
|
||
function togglePrediction() {
|
||
if (classifier.getNumClasses() === 0) {
|
||
alert("请先为至少一个手势采集样本后再开始预测！");
|
||
return;
|
||
}
|
||
isPredicting = !isPredicting;
|
||
updatePredictionUI();
|
||
}
|
||
|
||
/**
|
||
* 应用的主循环
|
||
*/
|
||
async function mainLoop() {
|
||
// --- 修改点 3: 使用 estimateHands 替代 estimatePoses ---
|
||
// flipHorizontal: false 确保模型输出的坐标与原视频方向一致 (非镜像)
|
||
const hands = await detector.estimateHands(videoElement, { flipHorizontal: false });
|
||
canvasCtx.clearRect(0, 0, canvasElement.width, canvasElement.height); // 清空画布
|
||
|
||
if (hands && hands.length > 0) {
|
||
// 通常只处理检测到的第一只手，如果有两只手，可以根据需求处理
|
||
drawHand(hands[0]); // 使用新的 drawHand
|
||
if (isPredicting && classifier.getNumClasses() > 0) {
|
||
const handTensor = flattenHand(hands[0]); // 使用新的 flattenHand
|
||
const result = await classifier.predictClass(handTensor, 3);
|
||
handTensor.dispose();
|
||
|
||
const confidence = Math.round(result.confidences[result.label] * 100);
|
||
const predictedClassName = appState.classMap[result.label] || '未知手势'; // 文案修改
|
||
resultElement.innerText = `手势: ${predictedClassName} (${confidence}%)`; // 文案修改
|
||
}
|
||
}
|
||
animationFrameId = requestAnimationFrame(mainLoop);
|
||
}
|
||
|
||
// --- 模型管理函数 (无需修改，与之前一致) ---
|
||
|
||
/**
|
||
* 导出KNN模型为包含类别信息的JSON文件
|
||
*/
|
||
function exportModel() {
|
||
if (classifier.getNumClasses() === 0) {
|
||
alert('模型中还没有任何样本，无法导出！');
|
||
return;
|
||
}
|
||
|
||
const dataset = classifier.getClassifierDataset();
|
||
const datasetObj = {};
|
||
Object.keys(dataset).forEach((key) => {
|
||
const data = dataset[key];
|
||
datasetObj[key] = data.arraySync();
|
||
});
|
||
|
||
const modelData = {
|
||
classMap: appState.classMap,
|
||
dataset: datasetObj
|
||
};
|
||
|
||
const jsonStr = JSON.stringify(modelData);
|
||
const blob = new Blob([jsonStr], { type: "application/json" });
|
||
const url = URL.createObjectURL(blob);
|
||
const a = document.createElement('a');
|
||
a.href = url;
|
||
a.download = `hand-knn-model.json`; // 文件名改为 hand-knn-model.json
|
||
document.body.appendChild(a);
|
||
a.click();
|
||
document.body.removeChild(a);
|
||
URL.revokeObjectURL(url);
|
||
}
|
||
|
||
/**
|
||
* 从JSON文件导入KNN模型并恢复类别状态 (无需修改，与之前一致)
|
||
* @param {Event} event
|
||
*/
|
||
function importModel(event) {
|
||
const file = event.target.files[0];
|
||
if (!file) return;
|
||
|
||
const reader = new FileReader();
|
||
reader.onload = (e) => {
|
||
try {
|
||
const modelData = JSON.parse(e.target.result);
|
||
|
||
if (!modelData.classMap || !modelData.dataset) {
|
||
throw new Error("无效的模型文件格式。");
|
||
}
|
||
|
||
classifier.clearAllClasses();
|
||
poseClassesContainer.innerHTML = '';
|
||
appState.classMap = {};
|
||
|
||
appState.classMap = modelData.classMap;
|
||
const classIds = Object.keys(appState.classMap).map(Number);
|
||
appState.nextClassId = classIds.length > 0 ? Math.max(...classIds) + 1 : 0;
|
||
|
||
classIds.forEach(id => {
|
||
createClassUI(id, appState.classMap[id]);
|
||
});
|
||
|
||
const newDataset = {};
|
||
Object.keys(modelData.dataset).forEach((key) => {
|
||
newDataset[key] = tf.tensor(modelData.dataset[key]);
|
||
});
|
||
classifier.setClassifierDataset(newDataset);
|
||
|
||
updateSampleCounts();
|
||
checkExportAbility();
|
||
alert('模型导入成功！');
|
||
|
||
} catch (error) {
|
||
console.error("导入模型失败:", error);
|
||
alert(`导入失败！请确保文件是正确的模型JSON文件。\n错误: ${error.message}`);
|
||
} finally {
|
||
fileImporter.value = '';
|
||
}
|
||
};
|
||
reader.readAsText(file);
|
||
}
|
||
|
||
|
||
// --- 辅助和UI更新函数 ---
|
||
|
||
/**
|
||
* --- 修改点 4: 展平手部关键点 ---
|
||
* 将手部关键点展平为一维张量。
|
||
* 考虑到 MediaPipe Hands 模型的关键点总数是21个 (0-20)。
|
||
* @param {Object} hand - 单个手部检测结果对象
|
||
* @returns {tf.Tensor} - 展平后的关键点坐标张量
|
||
*/
|
||
function flattenHand(hand) {
|
||
// 归一化关键点坐标到 [0, 1] 范围，然后展平
|
||
const keypoints = hand.keypoints.map(p => [p.x / videoElement.videoWidth, p.y / videoElement.videoHeight]).flat();
|
||
return tf.tensor(keypoints);
|
||
}
|
||
|
||
const HAND_CONNECTIONS = [
|
||
[0, 1], [1, 2], [2, 3], [3, 4], // Thumb
|
||
[0, 5], [5, 6], [6, 7], [7, 8], // Index finger
|
||
[0, 9], [9, 10], [10, 11], [11, 12], // Middle finger
|
||
[0, 13], [13, 14], [14, 15], [15, 16], // Ring finger
|
||
[0, 17], [17, 18], [18, 19], [19, 20], // Pinky finger
|
||
[0, 5], [5, 9], [9, 13], [13, 17], [17, 0] // Palm base connections
|
||
];
|
||
|
||
/**
|
||
* --- 修改点 5: 绘制手部骨骼 ---
|
||
* 绘制手部关键点和连接线。
|
||
* @param {Object} hand - 单个手部检测结果对象
|
||
*/
|
||
function drawHand(hand) {
|
||
if (hand.keypoints) {
|
||
const keypoints = hand.keypoints;
|
||
|
||
// 绘制连接线
|
||
canvasCtx.strokeStyle = '#00FFFF'; // 青色
|
||
canvasCtx.lineWidth = 2;
|
||
|
||
for (const connection of HAND_CONNECTIONS) {
|
||
const start = keypoints[connection[0]];
|
||
const end = keypoints[connection[1]];
|
||
// 检查关键点是否存在
|
||
if (start && end) {
|
||
canvasCtx.beginPath();
|
||
canvasCtx.moveTo(start.x, start.y);
|
||
canvasCtx.lineTo(end.x, end.y);
|
||
canvasCtx.stroke();
|
||
}
|
||
}
|
||
|
||
// 绘制关键点
|
||
canvasCtx.fillStyle = '#FF0000'; // 红色
|
||
for (const keypoint of keypoints) {
|
||
if (keypoint) { // 同样检查
|
||
canvasCtx.beginPath();
|
||
// 关键点半径设置小一点，因为手部关键点比人体姿态更密集
|
||
canvasCtx.arc(keypoint.x, keypoint.y, 4, 0, 2 * Math.PI);
|
||
canvasCtx.fill();
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
|
||
/**
|
||
* 更新所有类别UI上的样本数量 (无需修改，与之前一致)
|
||
*/
|
||
function updateSampleCounts() {
|
||
const dataset = classifier.getClassifierDataset();
|
||
const allClassElements = document.querySelectorAll('.pose-class');
|
||
allClassElements.forEach(el => {
|
||
const classId = parseInt(el.dataset.classId, 10);
|
||
const classInfo = dataset[classId];
|
||
// 确保 classInfo 存在，因为 classifier.clearClass(id) 后，dataset[id] 可能会是 undefined
|
||
const count = classInfo ? classInfo.shape[0] : 0;
|
||
el.querySelector('.sample-count').innerText = `(${count} 样本)`;
|
||
});
|
||
}
|
||
|
||
/**
|
||
* 根据状态更新UI (少量文案修改)
|
||
*/
|
||
function updatePredictionUI() {
|
||
const allActionButtons = document.querySelectorAll('.btn-sample, .btn-delete-class, .btn-add-class, #btn-import');
|
||
if (isPredicting) {
|
||
predictButton.innerText = "停止预测";
|
||
predictButton.classList.add('stop');
|
||
resultElement.innerText = "正在分析手势..."; // 文案修改
|
||
allActionButtons.forEach(btn => btn.disabled = true);
|
||
document.querySelectorAll('.class-name-input').forEach(input => input.disabled = true);
|
||
checkExportAbility();
|
||
} else {
|
||
predictButton.innerText = "开始预测";
|
||
predictButton.classList.remove('stop');
|
||
resultElement.innerText = "已停止";
|
||
allActionButtons.forEach(btn => btn.disabled = false);
|
||
document.querySelectorAll('.class-name-input').forEach(input => input.disabled = false);
|
||
checkExportAbility();
|
||
}
|
||
// 只有在有类别且有样本时才能预测
|
||
predictButton.disabled = isPredicting ? false : classifier.getNumClasses() === 0;
|
||
}
|
||
|
||
function enableControls() {
|
||
[predictButton, importButton, exportButton, addClassButton].forEach(btn => btn.disabled = false);
|
||
checkExportAbility();
|
||
}
|
||
|
||
/** 检查是否可以导出模型并更新按钮状态 */
|
||
function checkExportAbility() {
|
||
exportButton.disabled = isPredicting || classifier.getNumClasses() === 0;
|
||
}
|
||
|
||
// 释放 TensorFlow.js 相关的内存
|
||
function cleanup() {
|
||
if (detector) {
|
||
// 对于 MediaPipe runtime，detector.dispose() 可能不是必须的，
|
||
// 其内部会管理WebGL资源。但为保险起见可以保留。
|
||
// 或者更彻底地，如果不再需要，可以手动清理所有tf.Tensor。
|
||
}
|
||
if (classifier) classifier.clearAllClasses();
|
||
if (animationFrameId) cancelAnimationFrame(animationFrameId);
|
||
tf.disposeAll(); // 额外添加，确保所有创建的张量都被释放，防止内存泄露
|
||
console.log("Cleanup complete. All TensorFlow.js tensors disposed.");
|
||
}
|
||
|
||
// --- 启动应用 ---
|
||
window.onbeforeunload = cleanup; // 页面关闭前清理资源
|
||
init();
|