|
On the features of codec implementation
Let's talk about standard compression algorithms, without taking into account proprietary codecs independently developed by individual companies, the implementation features of which are unknown in most cases, and the advantages of their use can only be judged by the statements of the companies themselves.
Before discussing the pros and cons of using frame-by-frame and interframe compression, the following circumstance is worth mentioning. The standards of interframe compression algorithms most often used in CCTV, such as MPEG2/4, H.264, specify requirements for the structure of compressed data. At the same time, the method of calculating individual parameters (such as, for example, the displacement vectors of moving objects in intermediate frames) is left to the discretion of the developers of a specific implementation of the algorithm. Therefore, the quality of the implementation of interframe compression algorithms in video surveillance systems from different manufacturers can vary significantly. This is reflected in the quality of the image, the presence of various artifacts, as well as the volume of the compressed data flow.
At the same time, for frame-by-frame algorithms, such as JPEG, JPEG2000, the standard does not allow any liberties, and different implementations may differ only in performance.
Network traffic and archive size
The volume of traffic generated by transmitted video data compressed by interframe compression algorithms is significantly smaller than that of data compressed by frame-by-frame codecs. The same applies to the size of the video data archive. In fact, this is the point of using interframe compression.
For an image in full PAL resolution of 720 x 576 at 25 frames per second, the video data stream compressed by the JPEG2000 codec is about 12-16 Mbit/sec. With the same image quality, the stream compressed by MPEG2/4 is usually 4-8 Mbit/sec, depending on the degree of motion intensity in the frame. A decrease in the data volume cannot but affect the quality of video materials as a whole. The fact is that after decompression, most frames in interframe algorithms are derivatives of the so-called reference frames, which are compressed entirely. Derived frames are obtained artificially: by restoring moving objects of the previous (or previous and subsequent) frame using displacement vectors. Some features of the human eye's perception of a moving image are used. To understand what is meant, it is enough to recall the picture that was obtained if you pressed pause on an analog VCR while viewing a dynamic scene. In most cases, if you miss the reference frame, the image was very blurry in areas with moving objects. However, when playing the recording normally, the image looks fine.
It is these «smeared» frames that make up the majority of the archive recorded by a video surveillance system using interframe codecs. If you need to analyze an incident recorded by a video surveillance system, it may happen that you will have only 1-2, or perhaps no, reference frames suitable for analyzing the recording and identifying objects. Only frames of frame-by-frame compression algorithms or reference frames of interframe algorithms are accepted for forensic examinations. Reference frames are usually spaced 1-2 seconds apart, and the gaps between them are filled with derivative frames that are not suitable for analysis.
At the same time, in a video recording system with frame-by-frame compression, each frame is a separate snapshot and can be used for analysis. In addition, it can be said that the frequency of reference frames in interframe codecs is an adjustable value. However, when the frequency of reference frames increases to a value close to frame-by-frame codecs, the volume of network traffic (and data archive) increases and becomes higher than that of frame-by-frame codecs. In addition, frame-by-frame codecs based on wavelet transformation have a property that can be successfully used to scale the video stream and optimize network traffic. This will be discussed below.
Frame thinning
If the channel capacity is not enough to transmit data or the system cannot handle unpacking a large amount of data, some video information is lost. For a system using a frame-by-frame codec, it is possible to painlessly thin the stream, which is expressed in skipping some frames. In this case, the quality of individual frames does not suffer in any way, only the smoothness of the display of moving objects is lost.
The situation is completely different with interframe codecs. Since keyframes are the basis for a whole series of intermediate frames, if the data containing the keyframe is lost, a whole time interval of the video image is lost. The picture freezes and then continues from another place.
Video data stream scaling and network traffic optimization
Frame-by-frame compression algorithms, such as wavelet and JPEG2000, are quite demanding of computing resources. As a rule, they lose to interframe algorithms here, although with the proviso of the existing degree of freedom in the implementation of interframe codecs.
On the other hand, frame-based codecs based on wavelet transformation (wavelet, JPEG2000) have a convenient data organization structure that can be successfully used to optimize and scale streams. The fact is that in these compression algorithms, data on each frame has a block structure. In this case, each block contains image data detailing the data from the previous block. This allows you to get a compressed data stream with a lower resolution image from a compressed data stream, using a minimum of computing resources. Whereas in the case of using interframe codecs, the frame must be unpacked, reduced and packed again.
The compression algorithm used is one of the main characteristics of the video surveillance system. It can be said that for tasks solved by professional security systems, video surveillance systems with frame-based compression algorithms are more applicable at large and medium-sized facilities. This statement is based on the fact that each frame obtained with such a system can be used for analysis.
At the same time, systems with interframe algorithms have their advantages, which in some cases can be fundamental, especially for small objects, where the documentary nature of the recordings is not too important and there are no strict requirements for image quality. This is partly confirmed by the fact that most DVRs, specialized devices for small objects, use interframe codecs.
|
