随着深度学习推理技术的不断发展,让小型设备运行深度学习成为可能,阿里体育等IT大厂,推出的“乐动力”、“天天跳绳”AI运动APP,让云上运动会、线上运动会、健身打卡、AI体育指导等概念空前火热。那么,能否将这些在APP成功应用的场景搬上微信小程序,分享这些概念的红利呢?本系列文章就带您一步一步从零开始开发一个AI运动小程序,本系列文章将使用“AI运动识别”小程序插件,插件详情可以前往微信服务市场搜索相应插件。
一、骨骼图绘制原理
人体骨骼图的绘制,是通过在camera
组件上附一个同等大小的透明canvas
组件,在上面绘制关键点达到与人体图像重合的目的。
二、绘制代码
代码语言:html复制<template>
<view class="human-detection">
<camera id="preview" class="preview" :style="videoStyles" flash="off" :device-position="deviceKey"
resolution="high" frame-size="low" @initdone="onCameraReady">
</camera>
<canvas v-if="poseDrawEnabled" class="preview graphs" type="2d" id="graphics" :style="videoStyles"></canvas>
</view>
</template>
<script>
const AiSports = requirePlugin("aiSport");
const PoseGraphs = AiSports.PoseGraphs;
const humanDetection = AiSports.humanDetection;
export default {
data() {
return {
zoom: 1,
deviceKey: "back",
previewWidth: 480,
previewHeight: 640,
previewRate: 1,
frameWidth: 480,
frameHeight: 640,
status: 'unknown',
fps: 0,
poseFps: 0,
isHumanBody: false
};
},
computed: {
videoStyles() {
const style = `width:${this.previewWidth}px;height:${this.previewHeight}px;`;
return style;
}
},
mounted() {
this.autoFitPreview(480, 640);
this.initCanvas();
},
methods: {
autoFitPreview(width, height) {
const sifno = uni.getSystemInfoSync();
let rate = sifno.windowWidth / width;
this.previewWidth = width * rate;
this.previewHeight = height * rate;
this.previewRate = rate;
this.frameWidth = width;
this.frameHeight = height;
},
initCanvas() {
const that = this;
const query = uni.createSelectorQuery().in(that);
query.select('#graphics')
.fields({
node: true,
size: true
})
.exec((res) => {
if (utils.isEmptyArray(res))
return;
const canvas = res[0].node;
const ctx = canvas.getContext('2d');
const dpr = uni.getSystemInfoSync().pixelRatio;
canvas.width = res[0].width * dpr;
canvas.height = res[0].height * dpr;
ctx.scale(dpr, dpr);
that.canvas = canvas;
that.ctx = ctx;
that.poseGraphs = new PoseGraphs(ctx, canvas.width, canvas.height, 1);
that.poseGraphs.lineColor = "#FF8E148C";//线条颜色
});
},
async detection(frame) {
const human = await humanDetection.detectionAsync(frame);
//无结果
if (!human)
this.poseGraphs.clear();
else
this.poseGraphs.drawing(human.keypoints);
},
initVideo() {
if (this.camera)
return;
const that = this;
this.camera = new CameraDevice();
this.camera.onFrame = frame => {
that.fps = that.camera.fps;
//重新自适应
if (frame.width != that.frameWidth || frame.height != that.frameHeight) {
that.autoFitPreview(frame.width, frame.height);
that.initCanvas();
}
that.detection(frame);
};
}
}
}
</script>
<style lang="scss">
.human-detection {
width: auto;
height: auto;
.preview {
margin: auto;
width: 480px;
height: 640px;
}
.graphs {
position: absolute;
top: 0;
left: 0;
z-index: 9999;
box-shadow: 0 0 14.4928rpx #CCC;
background-color: rgba(0, 0, 0, 0.01);
}
}
</style>
三、注意事项
小程序的抽帧图像大小与camera
实时图像可能不一致(https://developers.weixin.qq.com/miniprogram/dev/component/camera.html#Bug-Tip),所以camera
和canvas
组件必须保持与帧图像保持同比缩放,否则可能导致骨骼与实时图像不一致。
下篇我们将为您介绍如何进行运动分析,敬请期待...