Two heads are better than one.
when security depends on eliminating unwanted alarms and detecting real threats
In an ongoing effort to replicate the human brain's analytical skills in recognizing actions and patterns, developers are always looking for new ways to improve the reliability and accuracy of video systems' detection of unauthorized intrusions. One new strategy is to combine a well-known video detection (VMD) algorithm with emerging software — video analytics technology. The result is a new class of detector that combines the results of two different analyzers, identifies threats more accurately and eliminates a large number of unwanted alarms.
Video motion detectors have been around for many years. Quality detectors have a long track record of being reliable in high security applications: prisons, power plants, airports, banks, defense facilities, etc. They are extremely effective at detecting motion in the areas they are intended for and very good at eliminating the effects of global changes in the frame, such as weather or changes caused by the movement of a mast or bracket with a video camera. Video analytics, on the other hand, is a relatively new technology that has been achieved by increasing the processing power of computers. This allows for the recognition and classification of different types of objects appearing in a video image. To understand why it makes sense to use the two technologies together, let's focus on the differences between the two technologies.
VMD (Video Motion Detector)
detect motion in predefined areas of the camera image and generate alarms. They work by analyzing the analog video signal from a CCTV camera, or more precisely, by analyzing the electrical representation of grayscale in the image and monitoring these changes in brightness or contrast. In a still camera image in a constant-light environment, this change can be inferred to be a motion detection result. However, VMDs are designed for outdoor situations where light levels and other conditions vary widely. Certain conditions must be met to eliminate soft changes in illumination while still allowing detection of potential threats.
Firstly, its sensitivity must be very high to allow detection of a low-contrast object in front of an equally low-contrast background, and at the same time the detector must be able to analyze each individual detection zone in the scene for subtle differences caused by smooth changes in illumination. Comparing the results across the entire image is what allows the VMD to analyze which changes are global and therefore not a threat, and which are localized and most likely to pose a threat.
Secondly, the VMD must compare and store information about a sequence of frames from the same camera quickly enough to ensure that a fast-moving target is captured within the operating cycle as it passes through the scene.
And third, the VMD should allow the user to set up small detection zones in the far field, which should be much more sensitive than those in the near field. This should balance the fact that any target naturally appears larger and causes a greater change in signal when it is close to the camera than when the target is in the far field.
Video Analytics
Video analytics is a technology of using intelligent software to filter and manage real-time CCTV video images. In this case, these are the pixels in the video image that need to be analyzed. The software continuously analyzes the scene and creates a basic base model for itself, constantly updating it to detect global changes such as weather and lighting. The current images are continuous compared to the updated background image model, and any unnatural changes compared to the model are a specific detection target for the user. The user can define target objects by their probable size, speed, direction of movement and position, and he can also associate these properties. The video analytics system can recognize such objects as «person» and «car», «object appearance» and «object disappearance», movements: «fast», «slow», «wrong direction of movement», separately or only when logically related to each other. This allows the system to raise an alarm when a person moves quickly or a car moves in the wrong direction, but ignore a car moving quickly or a person walking in the wrong direction. The user can also define different areas of interest in the image and associate them with different target features and raise an alarm in response to this action.
This algorithm creates and continuously updates, in self-training mode, a background scene model based on the pixel content of the entire image. This model records global changes in the frame. The algorithm settings provide for setting the size, speed, and direction of movement of recognizable objects. If changes are detected in any group of pixels in the current frame when compared with the background scene model, the system analyzes this group to detect signs of compliance with the parameters of predefined recognizable objects and generates an alarm according to the alarm settings in this zone.
Separately, but together
When both detection systems operate independently of each other to monitor the same image, their activity can be logically combined so that an alarm is transmitted to the operator only when precisely defined criteria for both systems coincide at the same time. This means, for example, that a bird that might trigger a VMD alarm because it meets the movement and location criteria for a VMD alarm does not meet the defined threat size criteria for a VA, and thus no alarm will be generated when the two systems operate together. On the other hand, when the wind blows leaves in a bush and deceives the VA into detecting human or vehicle movement, the VMD does not react and no alarm is generated.
A duplicated intelligent video analysis system for outdoor detection in a complex environment is already a reality today. The duplicated video detector has VA software and a built-in VMD, which can operate completely separately or in tandem. Thus, the user has the option of using the system in any environmental conditions.