This work presents a video-camera-based fire/smoke sensing technique for early warning in antifire surveillance systems. By exploiting R-CNN (Region Convolutional Neural Network), a detection technique is developed for the measurement of the smoke and fire characteristics in restricted video surveillance environments, both indoor (e.g. a railway carriage, container, bus wagon, homes, offices), or outdoor (e.g. storage or parking areas). The considered application scenario, to reduce costs, is composed of a single, fixed camera per scene, working in the visible spectral range already installed in a closed-circuit television system for surveillance purposes. The training phase is done with indoor and outdoor image sets, with both smoke and non-smoke scenarios to assess the capability of true-positive/true-negative detection and false-positive/false-negative rejection. To generate the training set, a Ground Truth Labeler app is used and applied to the open-access Firesense dataset, including tens of indoor and outdoor fire/ smoke scenes developed as the output of an FP7 project, plus other videos not publicly available, provided by Trenitalia during specific fire/smoke tests on railway wagons performed at their testing facility in Osmannoro, Italy. The achieved results show that the proposed R-CNN technique is suitable for the creation of a smart video-surveillance system for fire/smoke detection.