Akıllı Kamera Sistemleri için AI Entegrasyonu: Görüntü İşleme ile Nasıl Başlanır?

Katmera SBC'leri ile bilgisayarlı görü projelerinizi hayata geçirmenin pratik yollarını keşfedin.

AI Donanım Gereksinimleri

GPU Hızlandırması

Performans Optimizasyonu:

import cv2
import numpy as np
# GPU hızlandırma kontrolü
def check_gpu_support():
    try:
        # OpenCL desteği kontrolü
        print("OpenCV build bilgileri:")
        print(cv2.getBuildInformation())
        
        # GPU cihazlarını listele
        if cv2.ocl.haveOpenCL():
            print("✓ OpenCL desteği mevcut")
            cv2.ocl.setUseOpenCL(True)
        else:
            print("✗ OpenCL desteği yok")
    except Exception as e:
        print(f"GPU kontrol hatası: {e}")
check_gpu_support()

Bellek Optimizasyonu

Model Cache Stratejileri:

import tensorflow as tf
class ModelCache:
    def __init__(self, cache_size=512):
        self.cache_size = cache_size  1024  1024  # MB to bytes
        self.models = {}
    
    def load_model(self, model_path):
        if model_path in self.models:
            return self.models[model_path]
        
        # Model yükleme ve quantization
        interpreter = tf.lite.Interpreter(
            model_path=model_path,
            num_threads=4  # Quad-core optimizasyonu
        )
        interpreter.allocate_tensors()
        
        self.models[model_path] = interpreter
        return interpreter
# Kullanım
cache = ModelCache(cache_size=128)  # 128MB cache
model = cache.load_model("yolo_quantized.tflite")

Popüler AI Kütüphaneleri

OpenCV Kurulumu ve Optimizasyonu

# Sistem paketlerini güncelleyin
sudo apt update && sudo apt upgrade -y
# OpenCV bağımlılıkları
sudo apt install -y python3-pip python3-dev
sudo apt install -y libopencv-dev python3-opencv
sudo apt install -y libatlas-base-dev liblapack-dev libeigen3-dev
sudo apt install -y libgtk-3-dev libavcodec-dev libavformat-dev
sudo apt install -y libswscale-dev libv4l-dev libxvidcore-dev libx264-dev
# Python OpenCV kurulumu
pip3 install opencv-python==4.8.1.78
pip3 install opencv-contrib-python==4.8.1.78
# GPU desteği için ek paketler
pip3 install opencv-python-headless

Performans testi:

import cv2
import time
import numpy as np
def benchmark_opencv():
    # Test görüntüsü oluştur
    img = np.random.randint(0, 255, (1080, 1920, 3), dtype=np.uint8)
    
    # CPU vs GPU performans karşılaştırması
    iterations = 100
    
    # CPU testi
    start_time = time.time()
    for _ in range(iterations):
        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        blurred = cv2.GaussianBlur(gray, (15, 15), 0)
    cpu_time = time.time() - start_time
    
    # GPU testi (eğer mevcut)
    if cv2.ocl.haveOpenCL():
        cv2.ocl.setUseOpenCL(True)
        gpu_img = cv2.UMat(img)
        
        start_time = time.time()
        for _ in range(iterations):
            gray = cv2.cvtColor(gpu_img, cv2.COLOR_BGR2GRAY)
            blurred = cv2.GaussianBlur(gray, (15, 15), 0)
        gpu_time = time.time() - start_time
        
        print(f"CPU Süresi: {cpu_time:.3f}s")
        print(f"GPU Süresi: {gpu_time:.3f}s")
        print(f"Hızlanma: {cpu_time/gpu_time:.2f}x")
    else:
        print(f"CPU Süresi: {cpu_time:.3f}s")
        print("GPU desteği bulunamadı")
benchmark_opencv()

TensorFlow Lite Edge Optimizasyonu

import tensorflow as tf
def convert_model_to_tflite(model_path, quantize=True):
    # Keras modelini yükle
    model = tf.keras.models.load_model(model_path)
    
    # TFLite converter
    converter = tf.lite.TFLiteConverter.from_keras_model(model)
    
    if quantize:
        # INT8 quantization (daha hızlı çıkarım)
        converter.optimizations = [tf.lite.Optimize.DEFAULT]
        converter.target_spec.supported_types = [tf.int8]
        
        # Representative dataset (opsiyonel)
        def representative_data_gen():
            for _ in range(100):
                data = np.random.random((1, 224, 224, 3))
                yield [data.astype(np.float32)]
        
        converter.representative_dataset = representative_data_gen
    
    # Model dönüştürme
    tflite_model = converter.convert()
    
    # Quantized modeli kaydet
    output_path = model_path.replace('.h5', '_quantized.tflite')
    with open(output_path, 'wb') as f:
        f.write(tflite_model)
    
    return output_path
# Kullanım
quantized_model = convert_model_to_tflite('my_model.h5')
print(f"Quantized model kaydedildi: {quantized_model}")

MediaPipe Framework Uygulamaları

import cv2
import mediapipe as mp
class FaceDetectionSystem:
    def __init__(self):
        self.mp_face_detection = mp.solutions.face_detection
        self.mp_drawing = mp.solutions.drawing_utils
        self.face_detection = self.mp_face_detection.FaceDetection(
            model_selection=0, min_detection_confidence=0.5)
    
    def detect_faces(self, image):
        # BGR'yi RGB'ye çevir
        rgb_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
        rgb_image.flags.writeable = False
        
        # Yüz tespiti
        results = self.face_detection.process(rgb_image)
        
        # Sonuçları işle
        rgb_image.flags.writeable = True
        image = cv2.cvtColor(rgb_image, cv2.COLOR_RGB2BGR)
        
        if results.detections:
            for detection in results.detections:
                self.mp_drawing.draw_detection(image, detection)
                
                # Güven skoru
                confidence = detection.score[0]
                print(f"Yüz tespit edildi: %{confidence*100:.1f} güven")
        
        return image
    
    def run_camera(self):
        cap = cv2.VideoCapture(0)
        cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
        cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 480)
        cap.set(cv2.CAP_PROP_FPS, 30)
        
        while cap.isOpened():
            ret, frame = cap.read()
            if not ret:
                break
            
            # Yüz tespiti
            frame = self.detect_faces(frame)
            
            # FPS hesaplama
            cv2.putText(frame, f"FPS: {cap.get(cv2.CAP_PROP_FPS):.1f}", 
                       (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
            
            cv2.imshow('Face Detection', frame)
            
            if cv2.waitKey(1) & 0xFF == ord('q'):
                break
        
        cap.release()
        cv2.destroyAllWindows()
# Kullanım
detector = FaceDetectionSystem()
detector.run_camera()

Pratik Projeler

1. Nesne Tanıma Sistemi

import torch
import cv2
from ultralytics import YOLO
class ObjectDetector:
    def __init__(self, model_path='yolov5s.pt'):
        self.model = YOLO(model_path)
        self.classes = self.model.names
    
    def detect(self, image):
        # YOLO çıkarımı
        results = self.model(image, conf=0.5)
        
        # Sonuçları işle
        for result in results:
            boxes = result.boxes
            for box in boxes:
                # Koordinatlar
                x1, y1, x2, y2 = box.xyxy[0]
                confidence = box.conf[0]
                class_id = box.cls[0]
                
                # Çizim
                cv2.rectangle(image, (int(x1), int(y1)), (int(x2), int(y2)), (0, 255, 0), 2)
                cv2.putText(image, f'{self.classes[int(class_id)]}: {confidence:.2f}',
                           (int(x1), int(y1-10)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
        
        return image
    
    def real_time_detection(self):
        cap = cv2.VideoCapture(0)
        
        while True:
            ret, frame = cap.read()
            if not ret:
                break
            
            # Nesne tespiti
            frame = self.detect(frame)
            
            cv2.imshow('Object Detection', frame)
            
            if cv2.waitKey(1) & 0xFF == ord('q'):
                break
        
        cap.release()
        cv2.destroyAllWindows()
# Kullanım
detector = ObjectDetector()
detector.real_time_detection()

2. Yüz Tanıma Güvenlik Sistemi

import face_recognition
import numpy as np
import sqlite3
import pickle
class FaceRecognitionSecurity:
    def __init__(self, db_path='faces.db'):
        self.db_path = db_path
        self.init_database()
        self.known_encodings = []
        self.known_names = []
        self.load_known_faces()
    
    def init_database(self):
        conn = sqlite3.connect(self.db_path)
        cursor = conn.cursor()
        cursor.execute('''
            CREATE TABLE IF NOT EXISTS faces (
                id INTEGER PRIMARY KEY,
                name TEXT UNIQUE,
                encoding BLOB,
                created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
            )
        ''')
        conn.commit()
        conn.close()
    
    def register_face(self, image_path, name):
        # Yüz encoding'i hesapla
        image = face_recognition.load_image_file(image_path)
        encodings = face_recognition.face_encodings(image)
        
        if len(encodings) > 0:
            encoding = encodings[0]
            
            # Database'e kaydet
            conn = sqlite3.connect(self.db_path)
            cursor = conn.cursor()
            
            encoding_blob = pickle.dumps(encoding)
            cursor.execute('INSERT OR REPLACE INTO faces (name, encoding) VALUES (?, ?)', 
                          (name, encoding_blob))
            conn.commit()
            conn.close()
            
            # Memory'ye yükle
            self.known_encodings.append(encoding)
            self.known_names.append(name)
            
            print(f"✓ {name} başarıyla kayıt edildi")
            return True
        else:
            print("✗ Yüz bulunamadı")
            return False
    
    def load_known_faces(self):
        conn = sqlite3.connect(self.db_path)
        cursor = conn.cursor()
        cursor.execute('SELECT name, encoding FROM faces')
        
        for name, encoding_blob in cursor.fetchall():
            encoding = pickle.loads(encoding_blob)
            self.known_encodings.append(encoding)
            self.known_names.append(name)
        
        conn.close()
        print(f"✓ {len(self.known_names)} kayıtlı yüz yüklendi")
    
    def recognize_face(self, image):
        # Yüz tespiti
        face_locations = face_recognition.face_locations(image)
        face_encodings = face_recognition.face_encodings(image, face_locations)
        
        recognized_faces = []
        
        for face_encoding in face_encodings:
            # Karşılaştırma
            distances = face_recognition.face_distance(self.known_encodings, face_encoding)
            matches = face_recognition.compare_faces(self.known_encodings, face_encoding, tolerance=0.6)
            
            name = "Bilinmiyor"
            confidence = 0
            
            if True in matches:
                match_index = np.argmin(distances)
                if matches[match_index]:
                    name = self.known_names[match_index]
                    confidence = 1 - distances[match_index]
            
            recognized_faces.append({
                'name': name,
                'confidence': confidence,
                'location': face_locations[len(recognized_faces)]
            })
        
        return recognized_faces
# Kullanım
security = FaceRecognitionSecurity()
# security.register_face('person1.jpg', 'Ahmet')
# faces = security.recognize_face(image)

3. Trafik Analizi Sistemi

import cv2
import numpy as np
from collections import defaultdict
import time
class TrafficAnalyzer:
    def __init__(self):
        self.vehicle_count = 0
        self.speed_data = []
        self.tracking_history = defaultdict(list)
        self.vehicle_classes = ['car', 'truck', 'bus', 'motorbike']
    
    def calculate_speed(self, track_id, current_pos, timestamp, pixel_to_meter=0.1):
        history = self.tracking_history[track_id]
        history.append((current_pos, timestamp))
        
        # En az 2 nokta gerekli
        if len(history) < 2:
            return 0
        
        # Son 5 ölçümü kullan
        if len(history) > 5:
            history = history[-5:]
        
        # Mesafe ve süre hesapla
        start_pos, start_time = history[0]
        end_pos, end_time = history[-1]
        
        distance = np.sqrt((end_pos[0] - start_pos[0])2 + (end_pos[1] - start_pos[1])2)
        time_diff = end_time - start_time
        
        if time_diff > 0:
            # Pixel/saniye'den km/saat'e çevir
            speed_mps = (distance * pixel_to_meter) / time_diff
            speed_kmh = speed_mps * 3.6
            return speed_kmh
        
        return 0
    
    def analyze_traffic(self, video_path):
        # YOLO modeli yükle
        net = cv2.dnn.readNet('yolov4.weights', 'yolov4.cfg')
        
        cap = cv2.VideoCapture(video_path)
        
        while cap.isOpened():
            ret, frame = cap.read()
            if not ret:
                break
            
            # Object detection
            blob = cv2.dnn.blobFromImage(frame, 1/255.0, (416, 416), swapRB=True, crop=False)
            net.setInput(blob)
            outputs = net.forward()
            
            # Araçları tespit et
            vehicles = self.detect_vehicles(frame, outputs)
            
            # Tracking ve hız hesaplama
            for vehicle in vehicles:
                track_id = vehicle['id']
                position = vehicle['center']
                timestamp = time.time()
                
                speed = self.calculate_speed(track_id, position, timestamp)
                
                # Görselleştirme
                x, y, w, h = vehicle['bbox']
                cv2.rectangle(frame, (x, y), (x+w, y+h), (0, 255, 0), 2)
                cv2.putText(frame, f'ID: {track_id}', (x, y-30), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
                cv2.putText(frame, f'Speed: {speed:.1f} km/h', (x, y-10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
            
            # İstatistikler
            cv2.putText(frame, f'Vehicles: {len(vehicles)}', (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
            cv2.putText(frame, f'Total Count: {self.vehicle_count}', (10, 60), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
            
            cv2.imshow('Traffic Analysis', frame)
            
            if cv2.waitKey(1) & 0xFF == ord('q'):
                break
        
        cap.release()
        cv2.destroyAllWindows()
# Kullanım
analyzer = TrafficAnalyzer()
# analyzer.analyze_traffic('traffic_video.mp4')

Performans İyileştirme Teknikleri

Model Pruning ve Optimization

import tensorflow_model_optimization as tfmot
def optimize_model(model):
    # Pruning configuration
    pruning_params = {
        'pruning_schedule': tfmot.sparsity.keras.PolynomialDecay(
            initial_sparsity=0.50,
            final_sparsity=0.80,
            begin_step=1000,
            end_step=2000
        )
    }
    
    # Model pruning
    model_for_pruning = tfmot.sparsity.keras.prune_low_magnitude(model, pruning_params)
    
    # Compile
    model_for_pruning.compile(
        optimizer='adam',
        loss='sparse_categorical_crossentropy',
        metrics=['accuracy']
    )
    
    return model_for_pruning
# Kullanım
optimized_model = optimize_model(your_model)

Multi-threading Optimization

import threading
import queue
from concurrent.futures import ThreadPoolExecutor
class MultiThreadProcessor:
    def __init__(self, num_threads=4):
        self.num_threads = num_threads
        self.frame_queue = queue.Queue(maxsize=10)
        self.result_queue = queue.Queue()
        self.executor = ThreadPoolExecutor(max_workers=num_threads)
    
    def process_frame(self, frame):
        # AI inference burada
        result = your_ai_model.predict(frame)
        return result
    
    def worker(self):
        while True:
            try:
                frame = self.frame_queue.get(timeout=1)
                result = self.process_frame(frame)
                self.result_queue.put(result)
                self.frame_queue.task_done()
            except queue.Empty:
                continue
    
    def start_processing(self):
        # Worker thread'lerini başlat
        for _ in range(self.num_threads):
            self.executor.submit(self.worker)
# Kullanım
processor = MultiThreadProcessor(num_threads=4)
processor.start_processing()

Etiketler