Forest smoke detection method using random forest classifier method

PURPOSE: A forest fire smoke detection method using a random forest classification technique is provided to improve the accuracy of smoke detection by determining whether a candidate block is a block indicating smoke or not according to the probability distribution of classes. CONSTITUTION: A key frame is selected among frames including at least one set candidate block(S200). Multiple feature vectors are extracted from the candidate block by comparing the selected key frame with one or more preceding key frames(S300). Learning by different random forest is performed about the extracted feature vectors(S400). According to each generated class, two first cumulative probability histograms are generated(S500). According to a class having the highest probability among the classes comprising a generated second cumulative probability histogram, the sensing condition of the candidate block is determined(S700). [Reference numerals] (S100) Step of setting a first reference value for dividing candidate blocks; (S200) Step of comparing images between consecutively photographed frames, setting blocks whose number of pixels in which a motion can be discriminated is equal to or greater than the first reference value as candidate blocks, and selecting a key frame from the frames including at least one candidate block; (S300) Step of comparing a key frame with the preceding frame and extracting a time feature vector and a space feature vector from the candidate blocks; (S400) Step of learning the time and space feature vectors by different random forest algorithm; (S500) Step of generating two first cumulative probability histograms by accumulating probabilities outputted to a terminal node with respect to classes resulting from the learning; (S600) Step of generating a second cumulative probability histogram by averaging the two first cumulative probability histograms according to class; (S700) Step of determining the sensing condition of the candidate blocks according to the highest probability glass of the second cumulative probability histogram as a delay or non-delay block; (S800) Step of marking the delay block to be discriminated from the non-delay block