Video Analyzer with OpenCV

What it does

This project analyzes videos using common OpenCV techniques: color masking (e.g. isolate red/orange objects), frame differencing (movement vs. a reference frame), and blob detection (locate simple objects like a ball and get their coordinates). It reads a video from AWS S3, runs these three analyses in parallel, and writes three annotated output videos back to S3 (mask-only, difference, and keypoints with a circle around the detected object). It can run locally or as an AWS Lambda function triggered when a new video is uploaded to S3, with optional transcoding via Elastic Transcoder.

Context

Part of a startup project at 4Geeks for image identification and video analysis. The implementation and concepts were documented in the 4Geeks blog: Analyzing Videos With Python.

Use cases

Video analysis is used in security, sports, cinema, transport, health, and home automation. This code focuses on:

• Movement detection — What changed between the first frame and the rest (frame differencing).
• Color-based filtering — Isolate objects by color (HSV mask) and ignore the rest.
• Object detection and position — Find blob-like objects (e.g. a ball), get (x, y) and size, and draw them on the video.

Defining what you want to analyze (detect vs. track, how many objects, entry/exit rules) drives how you implement it; this project demonstrates a simple pipeline for detection and masking.

How it was implemented

• Mask: Convert each frame to HSV, apply cv2.inRange with lower/upper red (e.g. [0, 130, 75]–[255, 255, 255]), then cv2.bitwise_and so only the colored region is visible.
• Frame difference: Store the first frame, then cv2.absdiff(first_frame, frame) per frame to highlight what moved (e.g. ball vs. static background).
• Blob detection: cv2.SimpleBlobDetector with color filter (e.g. blobColor 255 for dark blobs). Detector returns keypoints (position and size); the code draws a circle around the first keypoint and writes a “keypoints” video.
• Pipeline: One video is read from an S3 URL; three VideoWriters produce red_*, diff_*, and keypoints_* outputs. These are uploaded to S3 and optionally sent to AWS Elastic Transcoder for transcoding.

How to run it

Local

• Create a virtual environment and install dependencies:
  pip install -r requirements.txt
• AWS credentials: set ACCESS_KEY_ID and SECRET_ACCESS_KEY in the environment, or use a local settings module with those variables.
• Run the analyzer (example):
  python analyzer.py
  The default in __main__ uses a fixed S3 bucket and key (blog4geeks, original/vid_test.mp4); change those in analyzer.py for your bucket and object key.

AWS Lambda

• Package: ./make.sh (or zip analyzer.zip analyzer.py lambda_function.py).
• Deploy the zip as a Lambda function and set the handler to lambda_function.lambda_handler.
• Configure an S3 trigger so that when a video is uploaded to the bucket, the event passes bucket and object key to the handler; the handler creates a Detector and runs detect_movement().

Dependencies: Python 3, NumPy, OpenCV (opencv-python, opencv-contrib-python). For Lambda, include these in the deployment package (or use a layer). For AWS features: boto3 (add to requirements.txt if missing).

Technologies used

Python, OpenCV (cv2), NumPy, AWS (S3, Lambda, Elastic Transcoder), boto3.

---

Startup project at 4Geeks for image identification and video analysis. See Analyzing Videos With Python for the accompanying blog post.