fix(facial-recognition) avoid image data conversion

TF 3.0 supports getting an ImageBitmap in fromPixels: https://js.tensorflow.org/api/3.0.0/#browser.fromPixels

react/features/facial-recognition/facialExpressionsWorker.js

@@ -58,7 +58,7 @@ onmessage = async function(message) {
     }
 
     case SET_TIMEOUT : {
-        if (!message.data.imageData || !modelsURL) {
+        if (!message.data.imageBitmap || !modelsURL) {
             self.postMessage({
                 type: FACIAL_EXPRESSION_MESSAGE,
                 value: null
@@ -72,7 +72,7 @@ onmessage = async function(message) {
             modelsLoaded = true;
         }
         faceapi.tf.engine().startScope();
-        const tensor = faceapi.tf.browser.fromPixels(message.data.imageData);
+        const tensor = faceapi.tf.browser.fromPixels(message.data.imageBitmap);
         const detections = await faceapi.detectSingleFace(
                 tensor,
                 new faceapi.TinyFaceDetectorOptions()

react/features/facial-recognition/functions.js

@@ -131,17 +131,8 @@ export async function sendDataToWorker(
         return;
     }
 
-    const canvas = document.createElement('canvas');
-    const context = canvas.getContext('2d');
-
-    canvas.width = imageBitmap.width;
-    canvas.height = imageBitmap.height;
-    context.drawImage(imageBitmap, 0, 0);
-
-    const imageData = context.getImageData(0, 0, imageBitmap.width, imageBitmap.height);
-
     worker.postMessage({
         type: SET_TIMEOUT,
-        imageData
+        imageBitmap
     });
 }