Author: admin

Development of “DEFOG. Fog removal” module is underway for intelligent software IVISET.

Designation: the image in front of the camera may be covered with fog. Due to this, the moving objects become less sharp, many details are lost and they can only be detected by their shape. In order to ensure accurate object identification, the function of fog (smoke) removal is used in such cases.

Problem solving method: A special filter enhances contrast and improves color rendering of incoming video image. This is achieved through construction and equalization of a histogram.

Works are underway for creating a no-interest zone hiding module.

Designation:

It is often the case when the camera’s image area catches objects that are of no interest, yet interfere with the operation of the video analytics modules. For example, these include trees and flying flags. In the process of surveillance, these objects will be displaced, thus creating motion, which should not attract attention of the security guard.

Respectively, all such zones that are of no interest to the operator are excluded.

This module will help avoid identification errors in such cases.

A new module is being developed for IVISET — «Spherical warp removal».

Designation: some types of cameras have a spherical warp problem due to the lens operation. The image from such a camera is heavily warped, which results in incorrect assessment of the size of the objects within the camera’s field of vision, as well as of the distance to them. Curves appear at the edges of the picture (frame), which, respectively, warps the image. These warps interfere with the operation of identification and tracking detectors. Warped shape of the object is the main reason of object identification errors.

Problem solving method: The warped frame edges should be equalized along the top and bottom frame boundaries, thus equalizing the image. The frame edges that fall beyond the equalization boundaries are cut off.

In the near future, IVISET software will be supplemented with a module that processes spherical images from Fisheye cameras and transforms it into a panoramic one or divides it into several standard image windows.

Designation: wide view angle (180°) allows doing with one camera instead of several cameras looking at different directions. The image from such a camera has a shape of a circumference, in the center of which the warp is minimal, but the farther from the center the greater.

Displaying on a small screen of a wide view angle image results in small size of important details, which naturally affects the operation of motion sensors, object and face identification systems etc., and the operator loses spatial reference (if such a camera is ceiling-mounted) and cannot correctly identify a certain person or event.

Problem solving method: the image from such a camera, depending on the option of its installation, is transferred to a special filter that projects it on a cylinder. During projecting, proportions of all objects are equalized. The resulting image becomes spatially oriented.

The main purpose of the image heat analysis module is ensuring adequate level of security. There are two activity identification methods: dynamic and static.

In the first case, a color scale is laid over the video, and each moving object leaves a “trail” that acquires colder hues with time. Presence of “hot” zones means that there was an activity in a set area recently, which allows detecting the moving object even if it has already left the camera’s vision field.

The second heat map-based detection method is accumulation of “heat indices” in certain parts of the image. Apart from ensuring security, the analysis of “cold” and “hot” zones allows optimizing the good’s layout in a shop, better arranging exhibits at an exhibition, defining popular areas of an object etc.

In IVISET software, the heat map operation mechanism is realized based on the difference of frames.

The face recognition module is usually used for the tasks of control and access management. Thanks to the use of a neural network, the highest accuracy of people identification is achieved when installing the camera at any angle, which is impossible in the most other systems.

The face detector operates as follows: the sensor, at regular intervals, scans the face motion area. For smoothness of displaying, face tracking is performed at set intervals.

This module is capable of solving complex video analytics situations. Consider a specific example.

In a shop it is necessary to perform motion detection recording and people counting, but there are some complexities: the shops ceiling is very low (2.3 meters), due to which fact it is impossible to mount the camera on the ceiling and direct it downwards — people’s images are too large (1 person almost occupies the whole frame).
Possible way out — installing the camera at an angle. Yet, the problems persist.

The shop’s door is made of glass and leads to a street with a driveway. Passing by cars and pedestrians are seen through it. Near the shop’s entrance there are trees that create continuous play of light during wind gusts. All these result in a lot of false triggering of a regular motion sensor. The situation is further aggravated by reflections from the glass and the mirror hanging on it. Hence, the use of a standard motion detector is also impossible.

We suggest record start and entering people counting by faces. This option is optimal since the people that entered the shop will leave it sooner or later. And if it is necessary to calculate the time of a person’s staying inside the shop, a second camera may be installed.

The people counting module is usually used in shops, supermarkets and malls. Using this function, it is possible to assess the efficiency of trade premises operation, determine the most advantageous zones of the site, reveal weak tenants and increase efficiency of advertising campaigns.

This module will simplify monitoring of the general people traffic, buyers counting, assessment of convenience of the outlet position and taking into account the change in the buyers’ demand.

The module is implemented as follows: the software identifies the fact of and counts crossings of the line by moving objects of classes “face” and “person” in the motion area. For accurate identification of the fact of crossing, tracking of identified object is used.

The people counting module is based on the line crossing detector. In a classic case, the detectors operation is fully dependent in how well the motion sensor (detector) has identified the area. The area size is influenced by shadows: a usual motion detector, without shadow removal, identifies the overall dimensions 2-3 times greater than the actual ones. Iviset eliminates this problem via shadow removal.

However, this counting method has a drawback: it is not universally applicable. For example, a person with a cart or a big bag will be counted as 2 persons (or more).
Specially for this, Iviset has an in-built identification of moving object types, and in this case only the objects of the set category are counted.

Module designation: the coding results in noise and interference on the image. The module helps remove noises and smoothen the image. This is an important function since on a noisy image the interference may become a source of false motions and, consequently, a source of false sensor triggering. Also, noise and interference hinder the operation of a tracker (the tracker may stop tracking the object), and complicate object identification (decreased identification accuracy.)

Problem solving method: The filter is designed for restoring the quality of the video taken from DVD (for example, when 4 hours of video are recorded on it), VideoCD, shot by a mobile phone or after video codecs DivX, XviD, H.26x (H.261, H.263, H.264). Neat blocking artefact removal is performed, which is reflected in splitting of compressed video frames into squares.

The filter automatically determines the blocking level in each part of the square, and the result stores maximum details of the original video. Special blocking criteria guarantee accurate operation of each component of pre-processing and deblocking in case of both strong and barely perceptible blocking. So, in the same movie, at a scene with slow motion (almost blocking-free) the filter will operate minimally, while at scenes with heavy motion it will work at its full capacity.