Improving Image Stabilization in Modern Surveillance Cameras

Abstract: This thesis investigates potential ways to improve the image stabilization in Axis Communications surveillance cameras. The current in-house developed stabilization algorithm, Axis EIS, stabilizes based on gyroscope measurements. We compare a purely gyroscope based stabilization algorithm with a gyroscope-accelerometer sensor fusion stabilization algorithm, image-based stabilization algorithms and stabilization algorithms combining image analysis and gyroscope measurements. Footage from surveillance cameras are stabilized using these different approaches and the algorithms are evaluated based on image quality as well as their implementation viability in a real-time stabilization pipeline running on the camera. The results indicate that a purely gyroscope-based stabilization algorithm is difficult to surpass in a real-time environment due to its good qualitative performance relative to its low computational complexity. Simple image-based techniques could complement a gyro-based stabilizer if sufficient feature point tracking speed is attained.