Storage is expensive. Selective compression allows you to optimize storage and lower costs. But how exactly does this happen?
Given the massive amount of video data generated every day from an increasing number of cameras, optimizing storage is essential. Selective compression is one of the most preferred methods for this. Simply put, it is a process where AI-enabled software “selects” areas that are important in a video and compresses them less than other unimportant areas.
But how does the system define important and unimportant areas? What if you wanted a detail from a seemingly unimportant area? Which verticals should choose this method for storage optimization? We recently asked these questions to Timo Sachse, Product Analyst EMEA at
Axis Communications.
Also read: Why cloud storage is essential to video surveillance
How does selective compression work?
Selective compression follows a simple logic. The advanced analytics built into the surveillance cameras dynamically view ‘blocks’ within the image in relation to two properties - is there ‘structure’ in the block (such as lines, edges, patterns, etc.) and is there any motion?
“This results in a matrix of four possible combinations - no structure and motion, structure and no motion, no structure and no motion, and both structure and motion,” Sachse explained. “Depending on the respective combination at any time, the dynamic software automatically assesses and decides whether the block can be compressed further, or whether that would lose critical details that might be of relevance for the operators.”
The most valuable image blocks are the ones that contain both motion and structure. A person walking by would be a good example. The blocks representing the person need to be preserved in high quality, and therefore image compression is not increased beyond the selected default compression value of the camera or stream.
“At the other end of the spectrum, there are static parts of the image with no structure – a blank wall on the side of a tall building, for example,” Sachse continued. “These parts of the image can be heavily compressed as they don’t contain any valuable information. That said, the dynamic nature of the software means that should motion take place in such an area, the compression levels are automatically altered to allow for highly detailed images to be captured.”
What if there is an important detail in non-selected areas?
This is a serious concern. But Sachse points out that the dynamic nature of the software means that the cameras are automatically reacting to changes in the scene all the time. If an area that has previously seen no movement and contains little structure suddenly features one or the other – or both – the camera will react by compressing the corresponding parts of the image less.
“As an example, imagine a camera focused on a blank wall,” Sachse added. “With no motion in the scene and no structure, the images will be heavily compressed to reduce the bitrate. Should someone appear in the scene, the camera will then reduce compression to capture forensic details. If that person then starts spraying graffiti on the wall, not only will that be captured in detail, but when they leave, the scene will now include structure – the graffiti – and therefore compress the part of the image with the graffiti less than before.”
Where does selective video compression work best?
The nature of scenes has an essential role in deciding the efficacy of selective video compression. The more details that change from one frame to the next one, the less efficient image compression works in general.
In fact, if a scene is permanently crowded – for example, an airport - video compression won’t be applied much at all as there’s constant motion and structure. If, on the other hand, the camera is monitoring a town square, where the device also captures one part of the sky, the camera could compress the image blocks that show the sky while capturing all the details of the busy town square.
“There are methods to keep the bitrate low also for very complex scenes with a lot of movement,” Sachse said. “But those methods (e.g., Maximum or Constant Bitrate Limitation, MBR or CBR) are sacrificing details to keep the bitrate low. This might be ok for a webcam, but a surveillance camera serves a different purpose. Hence the bitrate should be allowed to adjust to the level of complexity of the scene being recorded.”
Selective video compression: is it the right choice for you?
A crucial part for system integrators is the research. For most systems integrators, the best option would be to work with manufacturers to provide different methods and features to save storage. When considering which system to choose, flexibility and the option to adjust the system to real-life conditions are essential.
Storage (which is directly impacted by image bitrate) is costly. Thus, the general approach of many manufacturers in the industry is to keep bitrates down by using
maximum bitrate limitations (MBR), which apply a hard cap to the bitrate.
“Many manufacturers in the field force the cameras to stay under this bitrate limit, which allows a more accurate storage calculation when planning a solution,” Sachse said. “However, this also means that the delta between the set limit and actual bitrate used to capture images will cause the cameras to compress the footage, even if this means lower-quality images than might be needed.
Crucial forensic details essential for an investigation could be lost purely due to a bitrate limit set on the camera. Hence, integrators should stick to solutions that include smart and dynamic compression, that adjusts to the situation and captures everything while still keeping the bitrate low.
In general, it is recommended to judge a camera’s image quality when recording a complex scene instead of the footage created during quiet periods. The image quality differences that lie between the different bitrate philosophies of the manufacturers can be huge. Limited bitrates are a valid option, but not by default and not without a thorough per scene/installation analysis of the consequences.