<p>The process of improving the visual quality of the video is termed Video Stabilization (VS). The video quality is enhanced by diminishing the unwanted motions and shakiness in the video occurring owing to camera motions and other environmental conditions. However, the absolute relative depths during 3-Dimensional Video Stabilization (3DVS) were not focused in any of the prevailing works. Therefore, this paper proposes an efficient absolute and relative depths estimation-based VS system utilizing Glorothegonal weights with Stepky-hard-based Long Short-Term Memory (GSLSTM) and Bezier curve with Cholesrical-centric Kalman Filter (BCKF). Primarily, the input video is converted into frames. Next, the noises are removed. Afterward, the lens distortions that happen within the frames are compensated, followed by shadow removal. Then, by utilizing the Kullback-Leiblerakaike Divergence-based Gaussian Mixture Model (KLDGMM), the depth information of the frames is preserved. Thereafter, the absolute and relative depths are estimated; then, the features are extracted. After that, by using GSLSTM, the steady and unsteady video frames are classified. The motion is estimated for the unsteady frames. By utilizing BCKF, the dynamic motions are identified and handled. Lastly, to obtain the stabilized video, the steady and smoothened frames are synthesized. According to the analysis results, the proposed model is less error-prone by obtaining a Mean Absolute Error (MAE) of 0.042.</p>