Case study of a perimeter security deployment with multiple camera tracking video analytics

1. Infrastructure description

A single-storey brick house with a base 11 by 13 m and the attic at two levels, the total height of 13 m located at a distance of 3 m from the old wooden houses on a plot size of 31 to 56 meters. The plot surrounds a wooden fence height of 2.5 meters from the front and 2 meters from the neighbors. Around the house are garden trees and shrubs, only about 30 stands. The house is located in a climatic zone Moscow area zone.

Potential threats and challenges of video surveillance system

Prior to the deployment of video surveillance systems have been documented cases of illegal entry of people to the site over the fence from the front and neighbors into the home through a window to rob. The objectives of this surveillance system are:

1. automatic detection of unauthorized people in the protected area in the early stages, before people get into the house;
2. automatically generate an alarm signal is operational, the transfer of alarm messages with documentary images;
3. event driven video recording, the formation of the documentary archive event log (index), and fast search capabilities;
4. map position and motion paths of strangers on the map of protected areas in real time.

Supplemented by customer requirements:

1. high accuracy outdoor video analytics, not more than 1 false alarm per month with a probability of intruder detection of at least 0.99;
2. reliability must satisfy (1) under adverse conditions, surveillance (in the dark, in rain and snow);
3. remote video monitoring and archive storage outside the facility;
4. quality of event recording (at least 25 frames per second, resolution of at least 580 TV lines, no visible distortion compression);
5. good image detail, the person Height 180 cm must meet at least 60 TV lines (10% of the frame height).

3. Hardware

Table 1 shows the main equipment  selected for constructing of a system of the  street surveillance.

Table 1 List of major equipment

The most important aspects of the Guidelines for selecting and installing cameras for providing video analytics, weatherproof operation are listed below:

1. Between cameras with high resolution and low sensitivity and cameras with low resolution and high sensitivity, preference is given to the latter. Selected installation of more standard-definition cameras that provide a more effective work in adverse weather conditions, compared with a decision based on a smaller number of megapixel cameras within a fixed budget and requirements to detail the scene.
2. Selected camera with black and white sensor based on CCD (CCD) in order to achieve maximum sensitivity. Simultaneously, it is worth noting that the new color sensors, such as cameras MVK 8141ARVI, do not give this black and white camera sensitivity (0.02 lux stated by the manufacturer).

4. Camera placement

The initial placement of the camera was the proposal by experts and tested using the software package CAD  "Amulet". Testing results are shown on the pictures 1-8, where we used the following color designations:

• Purple clear - visibility zone cameras
• Yellow Transparent - the area selected for testing
• The Green Zone - the region observed a single camera
• Blue zone - the area is observed in more than one camera
• Red Zone - an area without supervision

Feature of the method "Amulet" is a statistical modeling by the Monte Carlo method by a pseudo-random distribution of violations in a given volume. The method allows to evaluate the effectiveness of coating chambers in the control area, to identify the dead band and duplication of cells. To test the selected area around the house with the linear dimensions of 2 and 3 meters respectively in height and width.

As a result of statistical modeling assessed the probability of intruder detection chamber in a single exposure (the capture of one frame). Estimating the probability was 0.92. The obtained value is calculated taking into account the direction and camera settings, given environmental conditions and characteristics of the offender.

If you are using analytics to interframe motion analysis, ie the ability to accompany the objects, the probability of detection increases significantly with a decrease in the probability of false alarm. For example, video analytics company Synesis " with the correct placement of the camera provides srednesezonnuyu probability of finding a person not less than 0.99 . Significant increase in the sensitivity of the detector determines that the information is collected over time based on processing of multiple frames. If the probability of detecting a person in one frame is 0.92, the probability of detection of 25 frames (the appearance of a person for 2 seconds at a frequency of handling 12.5 frames per second) to reach at least 0.99. During the movement, changing the contrast of the observed violator, as well as an opportunity to amplify the signal due to the correlation of several low-contrast areas on the motion path.

Picture 1. Cameras 1 and 2 placement

Picture 2. Cameras 3 and 4 placement

Picture 3. Cameras 5 and 6 placement

Picture 4. Cameras 7 and 8 placement

Picture 5. Modeling placement,  cameras 1-2

Picture 6. Modeling placement,  cameras 3-4

Picture 7. Modeling placement,  cameras 5-6

Picture 8.  Vulnerable area

5. Video managment system (VMS)

The video management system (VMS) is  "Intellect" of iTV. The main arguments for choosing this platform are:

1. support open standard network video ONVIF to integrate with the transmission part (including the broadcast of video and metadata);
2. ability to search the video archive using metadata and settings;
3. client for remote access from both the local network from the Internet.

As a turnkey solution with software "intelligence" used VideoServer iX, delivered by aggregator. The server consists of hardware DEPO Race S1 with an Intel Core i7 930 at a frequency of 2.80 GHz and 250 GB hard disks and 2000 GB SATA-II. Installed software includes the operating system Microsoft Windows 7 Professional, video surveillance platform "Intellect" 4.8.0.247 Service Device Pack version 3.1.9.293, Device Manager ONVIF version 0.1.1966. Setting out in accordance with the Administrator's Guide MagicBox device.

Platform "Intellect" has provided a video recording of events, intelligent search the archive, and remote access.

Device Manager ONVIF used for initial device configuration MagicBox, in particular, to calibrate its built-in video analytics.

6. Outdoor video analytics

Recognition of streaming video with the automatic formation of alarm messages is implemented based on hardware video analytics built into the device MagicBox. High precision detection of people in outdoor surveillance with noise of natural origin, such as uneven lighting changes, the movement of trees, rain and camera shake from the wind, served as the main criterion for choosing this device.

The introduction of a specialized video analytics to protect the perimeter has significantly reduced the number of false positives. For example, motion detector, built into the system "intelligence", in spite of the fine-tuning masks and thresholds of sensitivity (Picture 9), gave over 1,000 alarms each month at the facility. Simultaneously, the detector did not respond in 20-30% of cases, the appearance of people in the field of view (Picture 10). Skipping events leads to the fact that the video was not made and an alarm message is sent.

Device MagicBox, whose settings are presented in Picture 11-12, reduced the number of false positives to 1-2 in the first month in the facility in the absence of missed alarms. Picture 5 shows an example of detecting and tracking human MagicBox device in the case when the detector does not react, "Intellect." Annual statistics of true and false positives at the moment is going to objectively evaluate the effectiveness of smart  video surveillance system  during different  seasons.

Picture 10. Errors of the embedded detector in the "Intellect" system: false alarm from the motion of the branches (left) and missed the alarming situation in 37 seconds (right)

6. Оценка качества изображения

Pictures 14-22 compare the frames received from the cameras IAC-0842 ARP producted by  Byterg during daylight (in daylight) and in the dark (using IR illumination). MagicBox device captured the signal PAL, made H.264 compression and transmission of streaming video in accordance with standard ONVIF.

Picture 14.  Camera 1 in daytime (left) and during the night (right)

Picture 15. Camera 2 in daytime (left) and during the night (right)

Picture 16. Camera 3 in daytime (left) and during the night (right)

Picture 17. Camera 4 in daytime (left) and during the night (right)

Picture 18. Camera 5 in daytime (left) and during the night (right)

Picture 19. Camera 6 in daytime (left) and during the night (right))

Picture 20. Camera 7 in daytime (left) and during the night (right)

Picture 21. Camera 8 in daytime (left) and during the night (right)

Picture 22. Camera 9 in daytime (left) and during the night (right)

Conclusions on the quality of the image:

1. The picture quality matches the specs cameras. With an extended dynamic range sensor is a good study frames are stored in conditions of strong fluctuations of brightness in the bright sun, as well as with strong light sources in the field of view (Fig. 9).
2. In low light conditions without lights at night continued to observe the scene (Fig. 12, 14, left).Nevertheless, for these cameras is recommended to stop the additional IR illuminator.
3. Optimal setting of focus depends on the wavelength of light and different modes of operation for the bright and dark by about 10 meters (compare the area of focus for the left and right of Fig. 6-14). When the average tuning sharpness across the frame can be considered satisfactory.
4. Despite the hard wall mount home, camera and appropriate image shaking during strong wind gusts.However, this did not lead to false positives analytics, thanks to the digital stabilizer, embedded device MagicBox.

7. Multiple camera tracking (MCT)

Street surveillance cameras located around the perimeter of the building with a good overlap of zones of visibility that enables you to implement a multiple camerat tracking video analytics. This technology allows to:

1. improve the accuracy of detection and reduce false positives due to the correlation of metadata analytics adjacent chambers;
2. eliminate the repeated tripping video analytics in the transition from man of the observation of one camera to another camera surveillance zone;
3. display the whole trajectory of a man on a map of the protected object based on video analysis as soon as all the cells;
4. apply the rules to multi-chamber trajectory on the map for more accurate detection of human behavior and to search for events in the archives;
5. automatically selects the optimal angle of observation per person as it is moved from cell to cell.

Picture 23. Local trajectories of the movements of a man in the field of view of several cameras (6th, 3rd, 4th, 2nd and 1st) before their consolidation with the multicamera video analytics.

Picture 24.  The construction of a whole trajectory of the movements of a man on the map using the multicamera video analytics.

Pictures 23-24 are an example of how an algorithm for multi-tracking is used for unifying of five trajectories of the same object, obtained from five different cameras. This allowed us to reconstruct the trajectory of an object at a distance much greater than the field of view of one camera. Moreover, this association can initiate only one alarm event at a real object (instead of 5, which would have been generated if the multicamera analytics had not been used).

9. Conclusions

We have reviewed the case study of a perimeter security deployment with multiple camera tracking video analytics. Features of this solution are:

1. good image quality under adverse viewing conditions (bright sunlight, low light, rain, snow, fog);
2. Perimeter security video analytics automatically generates an alarm signal to the operational and metadata for fast search for video archive;
3. integration of devices and video surveillance systems MagicBox protocol ONVIF 1.02;
4. statistical modeling threats, and the calculation of the efficiency of the surveillance system based on the methodology "Amulet"
5. The multi-video analytics for display and analysis of the trajectories of human movement on the map.