High Definition Intelligent Network Video
EN
RU
Name:
Email:
Phone:
Message:

Megapixel video analytics in difficult surveillance conditions
Please, note that the article is automtically translated from Russian into English, so the translation may not be accurate.

Victoria Bulycheva and Nikolai Ptitsyn, LLC Synesis

High-definition (HD) support is now regarded as the main driving force behind the surveillance market, increasing the share of network cameras (IP) and reducing the proportion of analog cameras to standard definition (SD). Indeed, under favorable conditions of observation HD sensor improves image details, which increases the recognition accuracy and the range of video monitoring system. On the other hand, the HD cameras are often inferior to SD cameras in such important parameters as the sensitivity and dynamic range. Under adverse conditions, such as insufficient or uneven illumination, these parameters are more important than resolution, as well as for humans, and for video analytics. Let us consider the application of analytics in HD cameras, as well as related issues of improving the quality of HD-image.

CCTV video analytics is a youngand  ambitious technology. Systems analysis of video processing while not widespread, and after the implementation is often not met the expectations of the customer. On the other hand, continuous improvement of technology in terms of recognition accuracy and ease of use, as well as the widespread adoption of industry standards (ONVIF, PSIA) making video analytics more attractive to end users.

The growing popularity of megapixel cameras increases the relevance of video analytics. In the transition to HD mode multiplied the need for bandwidth network channels and disk space are often higher technically possible level. Particularly problematic expansion of communication channels of extended and / or remote sites. Video analytics eliminates the continuous broadcast video and go to the video recording of event. In other words, video analytics can automatically generate a control signal for the beginning and end of the broadcast HD video when an object of observation in the field of view. To implement this scenario video analytics should be built into the transmitting device, such as a camera or encoder (see the article).

The accuracy of HD video analytics

What are the advantages of megapixel video analytics? Obviously, high detail scenes using HD cameras increases the accuracy of detecting, tracking and classification purposes, such as in automatic mode by means of video analytics, as in "manual" observation operator.

In the task of protecting the perimeter, we can speak of a marked increase in the angle of coverage and / or the range of one camera (see the article). While the range of the camera SD is 40-90 meters depending on the quality of equipment and observing conditions, the range of HD cameras in good condition can be 150 meters.

The graph shows a typical dependence of the integral indicator of the accuracy of F 1 on the distance between camera and observed the object for camera standard (SD) and high definition (HD) for scripting "Protecting the perimeter in the sterile area." Green lines show the characteristics of the video, under favorable conditions of observation. Red - under adverse conditions, suggesting the presence of factors such as precipitation (rain, snow, fog), poor lighting, camera shake, the inhomogeneous density of the air due to heating of the surface contamination of the lens, the uneven illumination.

Solid linyami shows characteristics of the camera HD, dashed - Camera SD. Both cameras have identical hardware and software platform and configuration, but differ only in the sensors. As is evident from the graphs, under ideal viewing conditions, distance analytics HD is 120 meters and distance analytics SD - 80 meters with an accuracy of F 1 = 0.90. Range of 40-50 meters are virtually identical in both HD and SD systems for high precision F 1 = 0.99.

Under adverse conditions, video analytics HD behaves slightly worse than analyst SD because of the greater of noise. An important task of the developer megapixel camera - ensure high quality image on scenes with complex lighting.

Note that in problems of facial recognition go to HD mode provides a significant increase in accuracy.

chart-1

Typical dependence of the integral indicator of the accuracy of F 1 on the distance between the camera and nablyudaemymobektom for camera standard (SD) and high definition (HD).

Necessary conditions for achieving high range of the camera is the balance of its technical characteristics of the sensor, optics and computer parts. The transition to HD video analytics developer confronts new challenges. The tab describes the main problems associated implementation megapixel video analytics, and possible solutions.

Problems in the transition from SD to HD

Proposed solutions

Incoming data stream for video analytics HD (0.88 MP mode 720p, 1.98 megapixel mode 1080p) is 2-5 times higher than the flow of SD (0,40 MP mode PAL). At the same time increasing the number of objects analyzed, which leads to a nonlinear increase in computational complexity. These factors significantly limit the scalability of server-side analytics to 1-2 channels on a CPU core.
  1. Distributed computing between the transmitting and receiving parts CCTV with built analytics.
  2. Using new algorithms for video analytics with a managed resource use and deterministic processing time of one frame.

Zone control HD cameras can be much greater than that of the camera SD, which increases the demands on the dynamic range. In other words, at a greater area may be more znachitelny difference between the lightest and darkest levels of brightness. In this case, the sensors HD usually has a range of up to 70 dB, while the SD sensors up to 120 dB. Video analytics does not register objects in areas beyond the standard dynamic range.
  1. Using a new generation of sensors that combine megapixel resolution and wide dynamic range (WDR). In this video analytics should also support the regime WDR, that is, to handle more than 8 bits per pixel. In practice, this increases resource use video analytics in 2 times
  2. Using analytics WDR and / or algorithms for adaptive dynamic range compression, for example, Iridix. Application of these algorithms allows for compatibility with standard 8-bit video analytics.

HD sensors are often inferior to the sensitivity of sensors to SD several times because of smaller light-sensitive element. Video analytics does not register objects in low light.
  1. Using a new generation of sensors that combine megapixel resolution and high sensitivity.
  2. Automatic activation of spatial and temporal bininga (bining) pixels in low light.
  3. Using a highly sensitive detectors based on textural characteristics
  4. Increase in illumination power

In the wide-angle HD camera may be more significant geometric distortion of the image (such as "barrels") and larger-scale variation of objects. Such distortions significantly reduce the recognition accuracy.
  1. Improving camera calibration algorithms, the introduction of non-linear coordinate transformations from 2D to 3D space.
  2. Using multi-scale video analysis algorithms.

The range of variation of pixel size of objects varies more significantly from the foreground to the back.Unable to select the parameters of analytics for a uniform treatment of all frame
  1. The use of video analysis of multiscale methods
  2. Improving camera calibration algorithms for more accurate selection of the right scale for detecting, tracking and classification of objects.

Adaptive dynamic range compression

Algorithm Iridix, developed and patented by the British company Apical Imaging, uses orthogonal ritinomorficheskoe image conversion (orthogonal retina-morphic image transform, ORMIT). This method of dynamic range compression is different from the simpler analogues that for each pixel of the image used individual gamma-transformation in the tone and color space. The method uses the results of the research work of man, where retinal neurons to compress the recorded images before passing along the optic nerve to the visual cortex. Iridix algorithm is often combined with algorithms for noise suppression in spatial and frequency domains. This technology is now commonly used in cameras such manufacturers as Sony, Olympus, Nikon, but have not yet found wide application in the field of television.

Мегапиксельное видеонаблюдение на вокзале с неравномерным солнечным освещением: (а) исходное изображение

(а)

Мегапиксельное видеонаблюдение на вокзале с неравномерным солнечным освещением: (б) результат обработки при помощи технологии Iridix

(b)

Megapixel video surveillance at the station with the uneven solar illumination: (a) original image (b) the result of processing using the Irdix technology.

To receive feedback from the sensor HD, it is necessary that analytical processing frame was made in the original resolution. Most videoanaliticheskih systems perform image processing at a resolution of 320 x 240 pixels or less. Sharply question the increase in the analysis of images on the server, where many times the computational cost increases not only in video analytics, but also to decompress the stream. Thus, more reasonable to build power embedded video analytics, thereby avoiding redundant computation for decompression of the flow at high resolution.

An important advantage of megapixel video analytics "on board" is the ability to process the video to compress it.Even the most modern codecs compress the image with a uniform quality throughout the image. In practice, various areas of the frame to be materially different from the perspective of the importance of detail for video analytics. In the foreground detail is not as important than the far. Loss of detail (blurring) of the image in the spatial and temporal region leads to a decrease in sensitivity of the detectors on the distant frontiers, which lowers the effective range of the surveillance system. This lack of analytics server is manifested to a greater extent in the systems of HD, than in systems SD.

Business solutions overview

Are video analytics solutions available today? Several manufacturers have demonstrated high-definition video analysis at the shows, but on the full commercial introduction of yet, however.

Most companies offer HD camera with video analytics, the possibility of which does not match the resolution of the sensor. For example, a Texas company CoVi Technologies sells the camera with a 1 MP sensor and embedded videoanalikioy ObjectVideo, the processing flow of only 0.07 megapixel (320 x 240 pixels). Similarly, the camera family IQeye Pro Line videoanalikoy work with standard definition via: sys or Agent Vi. It is obvious that the range of automatic target detection in such systems significantly lags behind the optical performance of the camera.

Good megapixel camera, for example, Axis Q1755, equipped with the basic intelligence that solves a certain class of problems, such as detecting motion in an office or tampering alarm, but not suited for outdoor surveillance under the protection of the perimeter or public places. By the Axis cameras can be connected to the server videoanaliku, but, as noted above, it does not reveal the potential megapixel sensor due to distortions and high computational cost of decompression.

In April 2010 an American company VideoIQ demonstrated megapixel video analytics at the conference ICS West in Las Vegas. The company said it submitted an analyst consumes 8 times less computing resources that a similar system, which offers embedded processing stream 1080p in real time. On the other hand, on the basis of this demonstration is not to say that the analyst VideoIQ meet modern requirements: the developer, presented only a motion detector HD, while the most difficult parts - tracking (maintenance) of facilities and work with the ever-changing backdrop behind the scenes.

Israeli company ioimage integrate more complex video analytics in the three megapixel camera ioicam mmp100dn. In contrast, VideoIQ, the camera uses two signal processor, which allows for tracking and real disturbing scenarios: crossing the signal line, left the subject, the disappearance of the subject, etc. Unfortunately, the developer did not mention in the open-source resolution and range videonalitiki.

The Russian company has recently launched Byterg camera MVC-IP67-CMOS-1, 3Mp. The camera uses a powerful digital signal processor DaVinci TMS320DM6467 at a frequency of 729 MHz from Texas Instruments and megapixel video analytics Synesis. Byterg megapixel camera can monitor an area of 2-3 times higher in HD mode as compared with the regime of SD. The family of cameras Cisco 4000 Series HD IP Cameras built on the same processor TMS320DM6467, but uses a simpler videoanaliku ObjectVideo standard definition.

The Russian company Synesis develops hardware and software platform for intelligent camera high definition.Feature of this platform is a unique CMOS sensor that provides 1.2 megapixel video stream at 45 frames per second, and exceptionally wide dynamic range 120 dB, previously available only in the cells SD. Bit depth of the sensor is 20 bits per channel, and therefore applies intelligent technology dynamic range compression Iridix. In this video analytics running on the original resolution of the sensor. According to the developers, the technical characteristics of the camera will be ever closer to the physical limit.

In conclusion, the megapixel video analytics is a promising direction of development of surveillance (see article). It allows you to significantly improve the accuracy of recognition of situations and / or a range of security systems, as well as make better use of storage infrastructure and data transmission. However, the transition to megapixel analytics requires a fundamental processing of existing algorithms for video processing.Implementation videoanaliki HD not only by increasing the number of servers, the processor speed and memory.
 
© 2009 - 2010 Synesis