Abstract : Recognizing human-related events is a current area of study. There has been a lot of interest in real-time video event recognition. This is a difficult task due to the spatial and temporal variations in the frame. The proposed work is a real-time video event recognition approach that integrates detection probability, descriptive labels, and classification. This method for developing recognition strategies from event descriptors uses multi-decision fusion with probabilities (MDFP), where a consecutive winning event for 5 detections being chosen. In contrast to current models, this work is novel in that it uses real-time videos. The deep neural network architecture is used to train the Kinetic 400 dataset. Outdoor and indoor videos are provided for testing this model in real time. Regarding Top1 and Top5 accuracy as well as event recognition accuracy, this new model is more reliable than earlier models.