The Effect of the Coming Out of Multi Mega-Pixel Cameras on the Analysis of Image Content
Automated image content analysis has a growing importance in video surveillance systems. Less can be heard about the designability of systems based on video analytics. Which is the minimum resolution necessary for the analysis? What size of an area can be covered by one camera?
The fundamental goal of the video analysis is to help the analyser or to enable to realise applications otherwise unachievable (e.g. traffic count). Automated image content analysis may complete and make human surveillance even more reliable, or simply make it unnecessary in many cases.
Recognition-Based Designing for Video Analytics
The coming out of megapixel and multi-megapixel cameras makes the recognition-based design imperative in systems using video analytics. The considerably higher resolution of these cameras requires designing. This will assure to get a system appropriate for customer expectations and that the customer be fully aware of the capabilities of the video surveillance system even before the implementation, in the phase of designing.
The design process based on recognisability of systems using video analytics, fundamentally corresponds to that of the human surveillance systems.
Defining the width field of vision in case of video analysis
However there is some differences to be considered.
In comparison with the automated image content recognition, human observers may recognise even smaller objects. The reason for this is that human brain perceives seen sights more complexly, comprehending other factors as well.
If we increase the resolution of the sensor in a system without video analysis, that will increase the width of the field of vision as well, since the size of the smallest object recognisable by the observer person (PoH) is independent of the horizontal resolution of the sensor.
In systems with video analysis the merely increasing the resolution of the sensor means no increase of the width of the field of vision automatically. The size of the smallest recognisable target object (PoVCA) is depending of the algorithm applied and the achievable computing performance.
Well-designed video surveillance systems use the available computing capacity optimally, there is small reserve available. Using the same algorithm, processing the increased amount of pixel is only possible by decreasing the speed of video analysis (fps)or by increasing the size of the smallest recognisable target object.
This means in overall that merely increasing the resolution of the sensor itself is useless; the video analysis would mostly exploit the resolution of the original sensor only.
If we want to exploit even the possibilities of the multi-megapixel sensors during video analysis, we have to increase the computing capacity of the camera. The use of new, higher capacity processors or target hardware may assure the surplus for necessary computing performance.
| Object recognition by 1.3 megapixel camera | Object recognition by 9 megapixel camera |
 |
 |
width of target object: O = 0,5 m
size of the smallest recognisable target object in case of human surveillance: PoH = 10 pixels
size of the smallest recognisable target object in case of video surveillance: PoVCA = 24 pixels
horizontal resolution of sensor: PW = 1280 pixels
Specifying the size of the smallest recognisable target object is of fundamental importance in respect of possibilities and barriers of video analytics. In case the back cover of the product does not include this information, then it should be determined by series of measuring. The knowledge of this data will make the designing based on recognition of systems with video analysis possible.
