Disclosed are processes for the centralised monitoring of territories to recognize forest and surface fires. A swiveling and tiltable camera installed at a monitoring site supplies images of overlapping observation sectors. In each observation sector a sequence of images includes a plurality of images is taken, at an interval which corresponds to fire and smoke dynamics. An on-site image-processing software supplies event warnings with indication of the position of the event site in the analysed image. A total image and an image sequence with image sections of the event site are then transmitted to a central station and reproduced at the central station as a continuous sequence in quick-motion mode. Event warnings with relevant data are blended into electronic maps at the central station. Cross-bearing is made possible by blending event warnings from adjacent monitoring sites. False alarms are minimized by marking known false alarm sources as exclusion zones.