Abstract:

In order to overcome the shortcomings of conventional detectors and improve the detection rate of the current smoke detection algorithms, a new smoke detection approach based on motion block was proposed. Firstly, motion blocks were extracted by frame difference method. Then, candidate smoke regions were determined according to the area and the result of the motion estimation of every connected region. Lastly, the high/low frequency energy eigenvalues were obtained via discrete wavelet transformation and the movement maintenance eigenvalue was obtained according to the result of the estimated direction of movement. These eigenvalues were used to determine whether there is smoke. The experimental results prove that the proposed algorithm is robust, anti\|disturb. It could detect the appearance of smoke timely and prevent fire effectively.

Key words: smoke detection, motion detection, main motion orientation, motion block, wavelet transformation