בינה מאלכותית RB14-11 : זיהוי פנים של אדם בתוך תמונה

מחבר:admin
פורסם:ספטמבר 21, 2025
קטגוריה:רובוטרוניקס כללי
תגובות:אין תגובות

# -*- coding: utf-8 -*-
"""
Face recognition in a group photo using InsightFace.
Supports multiple reference people (Trump, Bill, Mark, etc.)
Adds a navy blue header on the result.
"""

import cv2
import numpy as np
from insightface.app import FaceAnalysis

# ======================================================
# 1) Load the InsightFace model
# ======================================================
app = FaceAnalysis(name="buffalo_l")
app.prepare(ctx_id=-1, det_size=(640, 640))  # CPU only

def cosine_similarity(a, b):
    return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))

# ======================================================
# 2) Load reference faces into an array
# ======================================================
reference_list = [
    {"name": "Trump", "path": r"d:\temp\trump1.jpeg"},
    {"name": "Bill",  "path": r"d:\temp\bill1.jpeg"},
    {"name": "Mark",  "path": r"d:\temp\mark1.jpeg"}
]

for ref in reference_list:
    img = cv2.imread(ref["path"])
    if img is None:
        raise FileNotFoundError(f"Reference image not found: {ref['path']}")

    faces = app.get(img)
    if len(faces) == 0:
        raise ValueError(f"No face detected in reference image: {ref['path']}")

    ref["embedding"] = faces[0].embedding

    x1, y1, x2, y2 = map(int, faces[0].bbox)
    cv2.rectangle(img, (x1, y1), (x2, y2), (0, 255, 0), 2)
    cv2.putText(img, f"{ref['name']} (100%)", (x1, y1 - 10),
                cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2)

    cv2.imwrite(fr"d:\temp\reference_{ref['name']}.jpg", img)
    print(f"Saved reference photo with box: d:\\temp\\reference_{ref['name']}.jpg")

# ======================================================
# 3) Load group photo
# ======================================================
group_img = cv2.imread(r"d:\temp\group_photo1.jpeg")
if group_img is None:
    raise FileNotFoundError("Group photo not found!")

faces = app.get(group_img)
threshold = 0.4

# ======================================================
# 4) Process faces in group photo
# ======================================================
for face in faces:
    emb = face.embedding
    x1, y1, x2, y2 = map(int, face.bbox)

    best_name = "Unknown"
    best_score = -1
    for ref in reference_list:
        sim = cosine_similarity(ref["embedding"], emb)
        if sim > best_score:
            best_score = sim
            best_name = ref["name"]

    percent = best_score * 100
    if best_score < threshold:
        best_name = "Unknown"

    label = f"{best_name} {percent:.1f}%"
    color = (0, 255, 0) if best_name != "Unknown" else (0, 0, 255)

    cv2.rectangle(group_img, (x1, y1), (x2, y2), color, 2)
    cv2.putText(group_img, label, (x1, y1 - 10),
                cv2.FONT_HERSHEY_SIMPLEX, 0.6, color, 2)

# ======================================================
# 5) Add small navy blue header
# ======================================================
header_text = "RoboTronix.co.il  050-6399-001"
navy_blue = (128, 0, 0)  # BGR format
cv2.putText(group_img, header_text, (20, 30),
            cv2.FONT_HERSHEY_SIMPLEX, 0.7, navy_blue, 2)

# ======================================================
# 6) Save and show result
# ======================================================
cv2.imwrite(r"d:\temp\group_result.jpg", group_img)
cv2.imshow("Group Photo Result", group_img)
cv2.waitKey(0)
cv2.destroyAllWindows()

print("Group result saved at: d:\\temp\\group_result.jpg")

100

101

# -*- coding: utf-8 -*-

"""

Face recognition in a group photo using InsightFace.

Supports multiple reference people (Trump, Bill, Mark, etc.)

Adds a navy blue header on the result.

"""

import cv2

import numpy as np

from insightface.app import FaceAnalysis

# ======================================================

# 1) Load the InsightFace model

# ======================================================

app = FaceAnalysis(name="buffalo_l")

app.prepare(ctx_id=-1, det_size=(640, 640)) # CPU only

def cosine_similarity(a, b):

return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))

# ======================================================

# 2) Load reference faces into an array

# ======================================================

reference_list = [

{"name": "Trump", "path": r"d:\temp\trump1.jpeg"},

{"name": "Bill", "path": r"d:\temp\bill1.jpeg"},

{"name": "Mark", "path": r"d:\temp\mark1.jpeg"}

]

for ref in reference_list:

img = cv2.imread(ref["path"])

if img is None:

raise FileNotFoundError(f"Reference image not found: {ref['path']}")

faces = app.get(img)

if len(faces) == 0:

raise ValueError(f"No face detected in reference image: {ref['path']}")

ref["embedding"] = faces[0].embedding

x1, y1, x2, y2 = map(int, faces[0].bbox)

cv2.rectangle(img, (x1, y1), (x2, y2), (0, 255, 0), 2)

cv2.putText(img, f"{ref['name']} (100%)", (x1, y1 - 10),

cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2)

cv2.imwrite(fr"d:\temp\reference_{ref['name']}.jpg", img)

print(f"Saved reference photo with box: d:\\temp\\reference_{ref['name']}.jpg")

# ======================================================

# 3) Load group photo

# ======================================================

group_img = cv2.imread(r"d:\temp\group_photo1.jpeg")

if group_img is None:

raise FileNotFoundError("Group photo not found!")

faces = app.get(group_img)

threshold = 0.4

# ======================================================

# 4) Process faces in group photo

# ======================================================

for face in faces:

emb = face.embedding

x1, y1, x2, y2 = map(int, face.bbox)

best_name = "Unknown"

best_score = -1

for ref in reference_list:

sim = cosine_similarity(ref["embedding"], emb)

if sim > best_score:

best_score = sim

best_name = ref["name"]

percent = best_score * 100

if best_score < threshold:

best_name = "Unknown"

label = f"{best_name} {percent:.1f}%"

color = (0, 255, 0) if best_name != "Unknown" else (0, 0, 255)

cv2.rectangle(group_img, (x1, y1), (x2, y2), color, 2)

cv2.putText(group_img, label, (x1, y1 - 10),

cv2.FONT_HERSHEY_SIMPLEX, 0.6, color, 2)

# ======================================================

# 5) Add small navy blue header

# ======================================================

header_text = "RoboTronix.co.il 050-6399-001"

navy_blue = (128, 0, 0) # BGR format

cv2.putText(group_img, header_text, (20, 30),

cv2.FONT_HERSHEY_SIMPLEX, 0.7, navy_blue, 2)

# ======================================================

# 6) Save and show result

# ======================================================

cv2.imwrite(r"d:\temp\group_result.jpg", group_img)

cv2.imshow("Group Photo Result", group_img)

cv2.waitKey(0)

cv2.destroyAllWindows()

print("Group result saved at: d:\\temp\\group_result.jpg")

Step 1: Load the InsightFace Model

app = FaceAnalysis(name="buffalo_l")
Loads the buffalo_l model (high-quality recognition model).

app.prepare(ctx_id=-1, det_size=(640, 640))
Prepares the model.

ctx_id=-1 means run on CPU.
det_size=(640, 640) sets the input detection resolution.

Step 2: Cosine Similarity Function

def cosine_similarity(a, b):
Custom function to measure similarity between two face embeddings.
It returns a value between -1 and 1.

1 means identical.
0 means unrelated.
Negative means opposite direction (no relation).

Step 3: Load Reference Faces

We store known people’s faces as embeddings.
Each person can have multiple images for better accuracy.

reference_list = [ { "name": "Trump", "paths": […] }, … ]
A list of people with their image paths.

For each image:

Open with cv2.imread(path).
Detect faces with app.get(img).
Extract embedding: faces[0].embedding.
Store it in ref["embeddings"].

This means "Trump" will have embeddings from trump1.jpeg, trump2.jpeg, etc.

Step 4: Open Video File

input_video = r"d:\temp\movie_group_photo2.mp4"
Path to input video (or 0 for webcam).

raw_output = r"d:\temp\output_raw.avi"
Temporary AVI file before compression.

final_output = r"d:\temp\output_result.mp4"
Final MP4 file, ready for WhatsApp.

cap = cv2.VideoCapture(input_video)
Opens the video.

fourcc = cv2.VideoWriter_fourcc(*'MJPG')
Codec for AVI (MJPEG is widely supported).

fps = cap.get(cv2.CAP_PROP_FPS)
Gets frames per second from input video.

width, height store video size.

out = cv2.VideoWriter(raw_output, fourcc, fps, (width, height))
Creates video writer object for saving AVI output.

Step 5: Process Video Frame by Frame

while True:
Loop through each frame.

ret, frame = cap.read()
Reads a frame. If ret is False, video ended.

faces = app.get(frame)
Detects all faces in the frame.

For each face:

Extract embedding.
Compare with all reference embeddings using cosine_similarity.
Keep the best match.
If similarity < threshold (0.4), label as "Unknown".

cv2.rectangle(frame, (x1, y1), (x2, y2), color, 2)
Draws a box around the detected face.

Green if recognized.
Red if unknown.

cv2.putText(frame, label, (x1, y1 – 10), …)
Writes the person’s name and confidence percentage above the box.

Step 6: Add Header

header_text = "RoboTronix.co.il 050-6399-001"
Adds your brand header on top of each frame.

Step 7: Save and Preview

out.write(frame)
Writes the processed frame to AVI file.

cv2.imshow("Video Face Recognition", frame)
Shows live preview.
Press q to quit early.

Step 8: Cleanup

cap.release(), out.release()
Releases video input and output.

cv2.destroyAllWindows()
Closes preview window.

Step 1: Load the InsightFace Model

Step 2: Cosine Similarity Function

Step 3: Load Reference Faces

Step 4: Open Video File

Step 5: Process Video Frame by Frame

Step 6: Add Header

Step 7: Save and Preview

Step 8: Cleanup

אולי תאהב/י גם

ערכת לימוד רובוטיקה

ביגוד אלקטרוני

תכנון מצלמה – פיתוח אלקטרואופטיקה – עדשה

כתיבת תגובה לבטל