The effect of the appearance of multi-megapixel cameras on the Video Content Analysis
The importence of the automatic Video Content Analysis is increasing in surveillance industry, otherwise a very few information can be found of the designability of systems based on video analysis. What is the required minimum resolution for the analysis? What is the size of an area that can be covered with a single camera?
The main goal of the video analysis is to facilitate the work of the operator or to realize applications that had been previously infeasible (e.g. traffic count). Automatic Video Content Analysis can be an additional help and make human surveillance more reliable, or even make it unnecessary in many situations.
Recognition based designing for video analysis
The appearance of high resolution megapixel and multi-megapixel cameras in surveillance systems using video analysis necessitates recognition based designing, which assures that the realized system will fit best with customer expectations. In case of applying system design the client knows the exact system capabilities already before implementation.
The recognition based design process for systems using video analysis, basically corresponds to that of surveillance systems using human detection.
Defining the width of field of view in case of video analysis
However there is some differences to be considered.
In comparison with the automatic video content recognition, human observers may recognize even smaller objects. The reason for that is because human brain perceives the sight more complexly, comprehending several factors.
If we increase the resolution of the camera sensor in a system without video analysis, that will increase the width of the field of view as well, since the size of the smallest object recognizable by the observer (PoH) is independent of the horizontal resolution of the sensor.
Merely increasing the sensor resolution in systems with video analysis doesn't increase the width of the field of view automatically. The size of the smallest recognizable target object (PoVCA) is depending on the applied algorithm and the achievable computing performance.
Well-designed video surveillance systems are using optimally the available computing capacity, therefore only a small reserve is available. Processing an increased amount of pixels while using the equivalent algorithm is only possible by decreasing the speed of video analysis (FPS) or by increasing the size of the smallest recognizable target object.
Summarizing this means that merely increasing the resolution of the sensor is useless, the video analysis would mostly utilize only the resolution of the original sensor.
If we want to exploit the possibilities of the multi-megapixel sensor during video analysis, we have to increase the computing capacity of the camera. Using new, higher capacity processors or target hardware may ensure the surplus for necessary computing performance.
|Object recognition by 2 megapixel camera||Object recognition by 15 megapixel camera|
width of target object: O = 0.5 m
size of the smallest recognisable target object in case of human detection: PoH = 10 pixels
size of the smallest recognizable target object in case of automatic detection: PoVCA = 24 pixels
horizontal resolution of the sensor: P W = 1280 pixels
To know the possible applications and limits of video analysis, it is essential to specify the size of the smallest recognizable target object. If the data sheet of the product does not include this information, then it should be determined by series of measurements. The knowledge of this data realizes the recognition based designing of a system with video analysis.