Vehicle video stabilization algorithm based on grid motion statistics and adaptive kalman filtering

Abstract

Abstract Owing to the impact of vibration on the carrier of a car-mounted camera, video images are shaking, resulting in decreased or failed recognition accuracy based on visual-target detection. To solve this problem, a video stabilization algorithm based on grid motion statistics and an adaptive Kalman filter is proposed. To satisfy the real-time and precision requirements of the vehicle video stabilization algorithm, the Oriented Fast and Rotated Brief (ORB) feature point detection algorithm was selected to detect and describe the obtained video frames. In addition, the accuracy of the motion estimate is increased by deleting the erroneous match points using an erroneous match-removal algorithm based on grid motion statistics (GMS). The matching accuracy of the GMS-based feature matching algorithm increased by 2.3% and 4.1% compared to conventional feature-matching algorithms based on scale-invariant feature transform (SIFT) and ORB, respectively. However, the matching time between adjacent video frames was reduced by 76% and 16%, respectively. Considering the possible jitter of the vehicle-mounted camera, an adaptive Kalman filtering algorithm was used to smooth the acquired motion trajectory and solve the problem of classical Kalman filtering being sensitive to the initial value. The mean Peak Signal-to-Noise Ratio (PSNR) after stabilization rose by 10.27 dB in comparison to the video stability before stabilization. Therefore, this algorithm exhibits good stability.