Video content is produced in vast quantities, and there are numerous tools available for easily tampering with or forging videos. The developed deep learning model will efficiently classify input videos with moving backgrounds as either original or forged. Car dashcam videos dataset is used for model training, validation and evaluation. The dataset consists of 4214 videos, with 80% (3372) of the videos used for training and validation, and the remaining 20% (842) used for evaluation, the training and evaluation dataset is divided into equal number of forged and original videos. The model requires less computational speed and memory usage as we utilized the GOP (Group of Picture) structure based technique rather than a pixel-based technique for detecting forgery in video. GOP is a group of successive frame types that follows a specific pattern. As we considered GOP to train our model with, so we extracted and stored GOP pattern of all videos. Then the model was trained with the extracted GOP sequences. During the training process, it has been observed that validation and training losses are gradually decreasing, at end of last epochs the validation loss was 0.058, while the training loss was 0.057. When evaluated on the test dataset, the trained model performed exceptionally well, achieving a recall rate of 96.9%, a precision rate of 98.5%, and an F1 score of 97.7%.