Fast tunnel monitoring video stabilization method based on improved ORB feature

The stable video stream is the precondition and foundation for the detection of abnormal events. Because of the disturbance of vehicle light and local light mutation in the tunnel scene, the existing video stabilization methods have the problems of real-time and robustness, which are still difficult to meet the application needs. This paper proposes a fast video stabilization method based on improved ORB feature. The proposed method distinguishes the light interference and non-interference regions based on the gradient direction invariance of tunnel lamp regions. Then the modified KSW entropy is applied to obtain dynamic ORB thresholds of different regions. Next the method utilizes the priori distribution of salient feature points to narrow the feature extraction region. More, it establishes affine transformation model with nearest neighbor matching and least-squares to realize global compensation of the tunnel jitter image. Finally, the experimental results indicate that the proposed method improves the performance of real-time and the fluency of video stream, which can meet the application requirements of tunnel monitoring well.