||Since the IoT (Internet of Things) and home security is more and more common as technology progresses, the security that can be provided by surveillance systems receives more and more attentions. However, the aging effects or process defects of the video processing related circuits in surveillance systems may result in errors such that the objects of interests are not easy to be identified and thus raises the security concern. On the other hand, surveillance systems usually have a large number of digital video sources. If the maintenance staff needs to check each video source manually to see if repair or replacement is required, the incurred cost may be unaffordable. Fortunately, the video errors may not necessarily fail the surveillance system. If the error is insignificant, the objects are still likely identifiable, and thus the error is acceptable. Therefore, how to evaluate the acceptability of video signals is one critical issue to be addressed, which is helpful to extend the life time of video processing circuits and thereby the surveillance system. The required maintenance costs can also be reduced.|
In this thesis, we proposed a no-reference error-tolerability test method for videos. In most of the video applications, especially for the surveillance system or videos that are provided by sensors, the reference (golden) video data are usually unavailable, which makes implementation of an error-tolerability test process quite difficult. No-reference test methods can allow us to evaluate video quality by analyzing only characteristics inside the video without reference videos. In the literature there have been several no-reference video quality assessment methods developed. However, these methods focus on dealing with only noises during transmission or compression of video signals. So far no work targets errors due to problems inside video processing circuits such as aging. We believe that this thesis is the first one that proposes a no-reference error-tolerability test method for erroneous circuits.
In order to evaluate error-tolerability of erroneous videos, we simulate all the possible artifacts and generate various erroneous videos. We also analyze the characteristics of these generated erroneous videos as well their qualities. The relationship between video quality and its characteristics is also investigated. Accordingly an efficient error-tolerability test method is proposed, which can accurately evaluate acceptability of errors in short time. Compared with the previous researches, the proposed can achieve more than 90% test accuracy, while that of the previous method is only about 80%. In addition, our execution time is three times as fast as the previous method.