Fire Detection Algorithms Using Multimodal ... - Bilkent University

CHAPTER 6. WILDFIRE DETECTION 89When a slow moving region is classified as a rising region, i.e., p 1 ≫ p 2 , D 3 (x, n)is close to 1. Otherwise, the decision function D 3 (x, n) is close to −1.6.2.4 Shadow Detection and RemovalShadows of slow moving clouds are major source of false alarms for video basedwildfire smoke detection systems. Unfortunately, shadows of clouds have verylow U and V values, similar to smoke regions due to wildfires.The decision function for shadow regions are defined based on the shadowdetection method described in [41]. Average RGB values are calculated for slowmoving regions both in the current and the background images. Let S(n) representa slow moving region in the image I at frame number n. The average colorvector,follows:c I,S ⃗ (n), of this region in the image I at frame number n is calculated asc⃗I,S (n) = 1 ( ∑A S(n)x∈S(n)r I (x, n),∑x∈S(n)g I (x, n),∑x∈S(n)b I (x, n)) (6.6)where A S(n) is the area of the slow moving region S(n), and r I (x, n), g I (x, n),and b I (x, n) are the red, green and blue channel values of the pixel at location xin the n − th image frame I. Similarly, average color vector,region in the background image, B, is calculated as follows:c B,S ⃗ (n) = 1 ( ∑A S(n)x∈S(n)r B (x, n),∑x∈S(n)g B (x, n),∑x∈S(n)c B,S ⃗ , of the sameb B (x, n)) (6.7)where r B (x, n), g B (x, n), and b B (x, n) are the red, green and blue channel values ofthe pixel at location x in the background image frame B at frame number n. Weused the background image B slow as the background image in our implementation.andIn shadow regions, the angle, θ(x), between the average color vectors,c I,S ⃗c B,S ⃗ , should be small and the magnitude of the vector in the current imageshould be smaller than that of the vector in the background image, i.e.,| c⃗I,S (n)| < | c B,S ⃗ (n)| [41]. This is because shadow regions retain the color andthe underlying texture to some extent.