- To build a light-weight real-time vandalism detector with Image processing techniques which can be run on small processors like Raspberry-pi.
- The input is real-time surveillance footage.
- The system should immediately alert the user if vandalism is detected on the property monitored.
- We read each frame of the video in a parallel pipeline different from the image processing pipeline (Increases Frames per second for real-time operation).
- We apply Image enhancement techniques for removing noise, and converting the frame to gray scale.
- To identify motion,we use Background Subtraction with Weighted Average.
- We detect contours if any in the image, and keep a contour counter.
- Once a buffer time and countour counter exceeds the threshold the frame is passed to detect Vandalism on the property.
- Sliding Window Approach is used to extract small windows which could potentially contain Vandalism.
- Every window is passed to SSIM and if the value is below a set threshold, alert is generated.
- Install Imutils, scikit-learn, numpy and OpenCV. (I suggest using Anaconda)
- After downloading this, Use command prompt to run
python vandalism-detector.py-
- For passing a video file, use
python vandalism-detector.py -v "path to video file" - For passing size of Monitoring Area, use
python vandalism-detector.py -a "Monitoring Area size ". Default size is 500. - If no arguments are passed, the program automatically uses Webcam as default video source.
- For passing a video file, use
-
PyImageSearch tutorials have been extremely useful for utilizing various python libraries.