fix for detection threshold

websnap/script/main.js

@@ -1,7 +1,6 @@
 /* main.js - websnap */
 import { HandLandmarker, FilesetResolver }
   from "https://cdn.jsdelivr.net/npm/@mediapipe/tasks-vision@latest";
-
 /* set html elements */
 const video   = document.getElementById("video");
 const canvas  = document.getElementById("canvas");
@@ -9,9 +8,9 @@ const overlay = document.getElementById("overlay");
 const octx    = overlay.getContext("2d");
 const ctx     = canvas.getContext("2d");
 const flash   = document.getElementById("flash");
-
 let lastShot = 0;
-
+let peaceFrames = 0;
+const PEACE_FRAMES_REQUIRED = 6; /* ~100 ms at 60 fps */
 /* function to request fullscreen */
 function requestFS() {
   const el = document.documentElement;
@@ -19,19 +18,16 @@ function requestFS() {
     ?.call(el);
 }
 requestFS();
-
 /* add listeners for fullscreen */
 document.addEventListener("click",    requestFS, { once: true });
 document.addEventListener("touchend", requestFS, { once: true });
 document.addEventListener("dblclick", requestFS);
-
 /* camera setup */
 const stream = await navigator.mediaDevices.getUserMedia({
   /* ideal res: 4096x2160 ; 4k */
   video: { width: { ideal: 4096 }, height: { ideal: 2160 }, facingMode: "user" }
 });
 video.srcObject = stream;
-
 /* import mediapipe for handgestures */
 const vision = await FilesetResolver.forVisionTasks(
   "https://cdn.jsdelivr.net/npm/@mediapipe/tasks-vision/wasm"
@@ -39,18 +35,20 @@ const vision = await FilesetResolver.forVisionTasks(
 const handLandmarker = await HandLandmarker.createFromOptions(vision, {
   baseOptions: {
     modelAssetPath:
-      "https://storage.googleapis.com/mediapipe-models/hand_landmarker/hand_landmarker/float16/1/hand_landmarker.task"
+      "https://storage.googleapis.com/mediapipe-models/hand_landmarker/hand_landmarker/float16/1/hand_landmarker.task",
+    delegate: "GPU" /* faster inference → more consistent detections */
   },
   runningMode: "VIDEO",
-  numHands: 1
+  numHands: 1,
+  minHandDetectionConfidence: 0.3, /* default 0.5 — loosens initial detection */
+  minHandPresenceConfidence:  0.3, /* default 0.5 — keeps tracking through partial occlusion */
+  minTrackingConfidence:      0.3  /* default 0.5 — less likely to drop the hand mid-gesture */
 });
-
 /* flash for when picture gets taken */
 function triggerFlash() {
   flash.classList.add("pop");
   requestAnimationFrame(() => requestAnimationFrame(() => flash.classList.remove("pop")));
 }
-
 /* capture a picture */
 function takePhoto() {
   triggerFlash();
@@ -62,7 +60,6 @@ function takePhoto() {
     saveAs(blob, "websnap-" + Date.now() + ".png");
   }, "image/png");
 }
-
 /* detect shit */
 function detectLoop(timestamp) {
   const dw = overlay.offsetWidth, dh = overlay.offsetHeight;
@@ -71,35 +68,46 @@ function detectLoop(timestamp) {
   }
   /* draw rect */
   octx.clearRect(0, 0, dw, dh);
-
   if (video.readyState < 2) { requestAnimationFrame(detectLoop); return; }
-
   const result = handLandmarker.detectForVideo(video, timestamp);
-
   if (result.landmarks.length) {
     const lm = result.landmarks[0];
-
     /* finger indexes, for peace sign */
-    const indexUp   = lm[8].y  < lm[6].y;
-    const middleUp  = lm[12].y < lm[10].y;
-    const ringDown  = lm[16].y > lm[14].y;
-    const pinkyDown = lm[20].y > lm[18].y;
-    const fingerGap = Math.abs(lm[8].x - lm[12].x);
-    const peace = 
+
+    /* hand scale: wrist(0) → middle MCP(9) distance in normalised coords.
+       normalising against hand size makes thresholds work at any distance */
+    const scale = Math.hypot(lm[9].x - lm[0].x, lm[9].y - lm[0].y);
+    const minBend = scale * 0.5;
+
+    const fingerUp   = (tip, pip) => (pip.y - tip.y) > minBend;
+    const fingerDown = (tip, pip) => (tip.y - pip.y) > -minBend * 0.3; /* relaxed */
+
+    const indexUp   = fingerUp(lm[8],  lm[6]);
+    const middleUp  = fingerUp(lm[12], lm[10]);
+    const ringDown  = fingerDown(lm[16], lm[14]);
+    const pinkyDown = fingerDown(lm[20], lm[18]);
+
+    /* gap normalised so it works at any distance */
+    const fingerGap = Math.hypot(lm[8].x - lm[12].x, lm[8].y - lm[12].y) / scale;
+    const peace =
       indexUp &&
       middleUp &&
       ringDown &&
       pinkyDown &&
-      fingerGap > 0.05;
+      fingerGap > 0.35;
 
-    if (peace && Date.now() - lastShot > 2500) { /* wait a bit */
-      takePhoto();
-      lastShot = Date.now();
+    /* require gesture to hold for several frames to kill noisy false positives */
+    if (peace) {
+      peaceFrames++;
+      if (peaceFrames >= PEACE_FRAMES_REQUIRED && Date.now() - lastShot > 2500) { /* wait a bit */
+        takePhoto();
+        lastShot = Date.now();
+        peaceFrames = 0;
+      }
+    } else {
+      peaceFrames = 0;
     }
   }
-
   requestAnimationFrame(detectLoop);
 }
-
 requestAnimationFrame(detectLoop);
-