Computing technology advancements enable vision-based monitoring systems to detect fires more effectively through transformer frameworks. The application of convolutional neural networks (CNNs) to computer vision tasks has shown promising results during the past several decades. CNNs face challenges in detecting long-range dependencies because convolution operations introduce inherent inductive biases. Vision Transformers (ViT) have proven to be a strong replacement for CNNs in vision tasks because they treat images as patch sequences while enabling pixel attention internally. This research paper shows how Vision Transformers function as dependable automated systems for fire detection and smoke identification and human presence monitoring by collecting spatial information across the entire image. The proposed method underwent strict testing on a standardized dataset for indoor fire detection to confirm both robustness and effectiveness of the framework. The experimental findings indicated that ViT had better performance in terms of its 94.81% accuracy and 94.67% precision and 98.40% recall and 96.47% F1-score and 93.64% mean Average Precision (mAP) that revealed its high capability in real-time activities.