3D Rainy Stereoscopic Video Stabilization Using Depth Estimation

The process of enhancing the video’s quality by removing unwanted effects of camera shakes and jitters is called Video Stabilization (VS). However, the 3-Dimensional (3D) rainy stereoscopic VS process was not concentrated on any of the prevailing research work. Therefore, in this framework, an effective 3D rainy stereoscopic VS with depth estimation and Shape Autotuning Liebovitch map Cheetah Chase Algorithm with Convolution Neural Network (SA-LmCCA-CNN) is proposed. Primarily, the input videos are converted into a number of frames. After that, by using Pairnorm L0 Gradient Minimization (Pn-LGM), the raindrops in each frame are removed. Later, the overlapping region and depth estimation are processed, and by using the Liebovitch map Cheetah Chase Algorithm (LmCCA), the energy function is diminished. Likewise, to mitigate the hallucination issue, a mesh is generated by utilizing Alternating Least Squares-Locally Constrained Representations (ALS-LCR). Then, from the hallucination-mitigated image and energy function minimized image, the feature points are extracted. Later, by employing SA-LmCCA-CNN, the stable and unstable frames are classified. If the frame is unstable, then the frame undergoes motion and camera path corrections, followed by raindrop reconstruction; otherwise, raindrop reconstruction is done directly for a stable frame. Lastly, in order to get the stabilized video, the frames are synthesized. The experimental analysis proved the proposed model’s robustness in 3D rainy stereoscopic VS by attaining a stability score of 0.93.