Indoor Video Smoke Detection Based On Image Processing

Intelligent video surveillance is highly considered for its broad prospect of application and potential economic value in the modern security systems. It extracts sensitive information automatically and sends warning signals without human intervention. This article breakthrough the traditional fire detector limits, and provides an indoor fire smoke method based on the digital video stream.In this paper, suspicious smoke area is detected by the combination method of frame difference and background difference base on the threshold segmentation of OTSU. Considering the complexity of multiple moving objects in indoor environment, a region segmentation method based on approximate connected area is proposed. This method is used to replace the time-consuming processing introduced by striking the largest connected region with the traditional morphological computation. Even when the smoke is closed to other moving objects, partial independent image blocks could still be segmented, so that the real-time and reliability of region extracting are enhanced successfully.Discrimination of smoke characteristic consists of three modules. At first, color measurement method based on the brightness and saturation is obtained by the analysis in the decomposition of the smoke block to different spaces. Secondly, energy measurement method based on wavelet coefficients is achieved through transforming smoke image. Finally, structural similarity measurement is proposed in respect that the smoke has the high correlation in local region, because smoke is well-distributed and consistent in the small area. The structure similarity could be fairly described the problem. Moreover, the final recognition result of smoke is determined by synthetic statistics the time axis and image local space of smoke image blocks meeting the three criterions.Therefore, the innovation of the work of this article can be listed as follows: 1. Image block extraction method based on approximate connected area. 2. Structure analysis method based on historical information. The combination of them improves the effectiveness of smoke detection under complex environment.The smoke detection algorithm was tested in the real-time monitoring environment and public database of smoke videos on the PC platform. It performed an effective and accurate recognition in different light intensity, smoke concentration, shoot distance and monitor scenes. Therefore, recognition of indoor smoke has been commendably realized by the algorithm presented in this paper.