Real-time smoke removal for the surveillance images under fire scenario

When a fire happens in a building, internal closed-circuit television system becomes less effective under the influences of hot smoke. The fire scenario is very similar to the environments with bad weather conditions such as haze, rain, and snow. Comparing those bad weather conditions, the fire scenarios are much complicated with difficulties in processing the images. This can be reflected by two important aspects: The lighting condition changes frequently inside the building, and the smoke is always in black while the particles under bad weather conditions are generally white. So a fast image restoration method (GL-MSR method) based on the multi-scale Retinex (MSR) was developed in this study to improve the detection accuracy under complicated fire or the similar situations. For the proposed GL-MSR method, the Gaussian pyramid was used to replace the Gaussian convolution where a lookup table was built to reduce the calculation time of the logarithmic algorithm. Compared with the traditional methods such as histogram equalization, the GL-MSR method shows a better result than the others and its operation time was found only 198 ms, almost 1/26 of the traditional processing time.