Research On Wide Dynamic Range Imaging For Video Surveillance Application

The dynamic range of a digital image refers to the ratio of the maximumpixel value and minimum pixel value. Intensity values of scenes in the realworld usually have a very broad dynamic range while the dynamic range oftraditional image capture device is limited due to its physical properties.When capturing a scene which contains a dynamic range that exceeds that ofthe camera, there will be a loss of detail in either the lowlight areas, thehighlight areas, or both. Currently studies on surveillance image acquisitionmainly focus on the spatial resolution. Acquisition speed and datacompression ratio are hot issues in the research on video system.Unfortunately, problems caused by the gap between dynamic range oftraditional camera and that of the real world are ignored. This articleintroduces multi-exposure based Wide dynamic range(WDR) technique to fillthis gap for indoor video surveillance application. After analysis on some key points of WDR, a solution is proposed.Currently, time domain multiple exposures based WDR technique is themost reliable method, which is adopted and studied in this article. Expandingdynamic range is achieved by post processing so that no extra hardwareinstruments are needed when using this method. There are two majordifficulties when applying wide dynamic technology in video area. Due to thelimited number of exposures, exposure time for each exposure should beselected reasonably and effectively to ensure high imaging quality. On theother hand, simple and efficient algorithm is in great need due to the highcomplexity of existing wide dynamic image synthesis algorithms.In this paper, two proposals are introduced to overcome thesedifficulties.This paper analyzes the tradeoff between dynamic range coverage anddetail preservation when the number of exposures is limited. We proposed acontent-based exposure control method which consists of regionsegmentation, scene classification and exposure strategy selection. Region ofinterest has a higher priority in this method, which meets the demand ofindoor surveillance application. In order to segment ROI from input image,an algorithm combining coterminous frame differencing and multi-thresholddetection is proposed under single exposure mode, while under multiple exposures mode, background differencing is utilized and the background issynthesized from radiance map.An exposure fusion algorithm in YUV color space is also proposed inthis article to achieve low computation complexity and high imaging quality.Guided by two measure factors, Y channel of input images are fused based onLaplacian pyramid. Maximum U and V values of input images are selected asthe UV components of the fused image. The proposed algorithm achievescomparable and even better detail and color preserving capacity according tothe experiment result.