Method and system for automatic forest fire recognition

摘要

A reference image of a scene is acquired and compared to the current image, using normalization, filtering, matching, cluster search algorithm and probability assessment, with triggering of an alarm if a smoke probability for a cluster exceeds a predefined threshold. The method involves using an automatic optical detector rotatably mounted on a platform, an electronic evaluation unit and a transmitter or local alarm. A reference image of a scene is acquired and the horizon determined. The image is normalized and the image region below the horizon marked. Non-linear filtering of the reference image suppresses any movement and the image is temporarily stored. A current image is acquired and matched to the reference image. The image is normalized as above. Thresholds are formed proportional to normalized standard deviations. The images are compared to produce a binary difference image using the thresholds. A cluster search algorithm is used to locate related areas. Probabilities are formed to assess the located clusters and eliminate those below a probability threshold. An alarm is triggered if the smoke probability for at least one cluster exceeds a predefined threshold. An Independent claim is also included for an arrangement for automatic forest fire detection.