OpenCV (Open Source Computer Vision Library) is an extensive collection of algorithms for real‑time image & video processing. Python bindings (cv2 module) offer a clean, readable API that integrates seamlessly with the broader NumPy / SciPy ecosystem.

Note ▸ Use the opencv-python wheel for most projects. The opencv-contrib-python wheel includes extra “contrib” modules (e.g. SIFT, SURF, aruco).

# Terminal — latest stable OpenCV 4.x
pip install opencv-python
# Extra modules
pip install opencv-contrib-python
            

2.1 Images

• cv2.imread(path, flags) — load BGR image.
• cv2.imwrite(path, image) — encode & save.

import cv2

img = cv2.imread("lenna.png", cv2.IMREAD_COLOR)   # BGR 8‑bit
cv2.imshow("Lenna (BGR)", img)
cv2.waitKey(0)
cv2.imwrite("lenna_gray.jpg",
            cv2.cvtColor(img, cv2.COLOR_BGR2GRAY))

2.2 Video Capture & Write

• cv2.VideoCapture(src) — camera / file reader.
• cv2.VideoWriter(outPath, fourcc, fps, size)

import cv2, time

cap   = cv2.VideoCapture(0)                       # webcam 0
four  = cv2.VideoWriter_fourcc(*"mp4v")
size  = (int(cap.get(3)), int(cap.get(4)))
out   = cv2.VideoWriter("capture.mp4", four, 30, size)

while cap.isOpened():
    ret, frame = cap.read()
    if not ret: break
    out.write(frame)          # save
    cv2.imshow("Live", frame) # preview
    if cv2.waitKey(1) == 27:  # ESC ↯
        break

cap.release(); out.release(); cv2.destroyAllWindows()

• cv2.resize(img, dsize[, fx, fy])
• cv2.cvtColor(img, code)
• img[y1:y2, x1:x2] — NumPy crop.

gray   = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
small  = cv2.resize(img, (0, 0), fx=0.25, fy=0.25,
                    interpolation=cv2.INTER_AREA)
face   = img[80:260, 70:210]  # simple crop

• cv2.rectangle(img, pt1, pt2, color, thk)
• cv2.circle, cv2.line, cv2.putText

annot = img.copy()
cv2.rectangle(annot, (70, 80), (210, 260), (0,255,0), 2)
cv2.putText(annot, "Face", (70, 75),
            cv2.FONT_HERSHEY_SIMPLEX, 0.7,
            (0,255,0), 2, cv2.LINE_AA)

Haar cascades are classic but still useful for low‑power demos. A pre‑trained XML file is shipped with OpenCV.

import cv2

detector = cv2.CascadeClassifier(
           cv2.data.haarcascades + "haarcascade_frontalface_default.xml")
cam = cv2.VideoCapture(0)

while True:
    ret, frame = cam.read()
    if not ret: break
    gray  = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    faces = detector.detectMultiScale(gray, 1.3, 5)

    for (x,y,w,h) in faces:
        cv2.rectangle(frame, (x,y), (x+w,y+h), (255,0,0), 2)

    cv2.imshow("Haar Face ↯ ESC to exit", frame)
    if cv2.waitKey(1) == 27:
        break
cam.release(); cv2.destroyAllWindows()

For modern accuracy‑speed trade‑offs choose a YOLO model. Either use Ultralytics pip install ultralytics wrapper or load ONNX weights directly with cv2.dnn.readNet.

from ultralytics import YOLO                                   # pip install ultralytics
import cv2, numpy as np

model  = YOLO("yolov8n.pt")                                    # nano model < 5 MB
cap    = cv2.VideoCapture(0)

while True:
    ok, frame = cap.read()
    if not ok: break
    results = model(frame, imgsz=640)[0]                       # predict
    annotated = results.plot()                                 # draw boxes
    cv2.imshow("YOLOv8 Live", annotated)
    if cv2.waitKey(1) == 27: break
cap.release(); cv2.destroyAllWindows()

Note ▸ OpenCV’s cv2.dnn can import ONNX, TensorFlow, Caffe, and Darknet formats with the same API.

• Official docs: docs.opencv.org
• Tutorial series: Python Tutorial Index
• Project ideas: barcode scanner, lane detection, AR marker tracking, edge‑AI with Raspberry Pi + NCS2.