A brief discussion on the five mainstream technologies of surveillance cameras

Wide dynamic range

Digital wide dynamic range does not achieve the purpose of expanding the dynamic range of imaging in a real sense, but it improves the contrast of local areas through the software image post-processing algorithm, which is generally achieved by the camera ISP module. The grayscale range that our naked eyes can distinguish is very limited, but in fact, computers can distinguish very weak grayscale differences. Digital wide dynamic range enhances these weak differences to the point that the naked eye can distinguish them through image processing algorithms. Later, dual-frame synthetic wide dynamic range appeared based on CCD hardware technology. The solution is to use a CCD, but each point on it is exposed twice in a single time, one long exposure (low shutter) and one short exposure (high shutter).

Therefore, each point has two data outputs, which is called "dual-output CCD". Using the DSP's unique image processing algorithm, the parts with appropriate brightness in the two images are cut out separately, and finally superimposed and synthesized to output an image with clear light and dark areas. Whether it is digital wide dynamic or dual shutter wide dynamic, the wide dynamic effect is not ideal.

With the evolution of DSP and CMOS technology, DPS uses a separate exposure and control technology for each pixel, and uses the linear superposition of multiple frames collected by the CMOS sensor to synthesize a complete image, which has a higher dynamic range than the double exposure imaging of CCD. From a numerical point of view, the dynamic range of CMOS cameras using DPS technology can reach 120dB or even 140dB based on the current processing technology. Wide dynamic technology has become an important indicator to measure the performance of a camera. At present, it has become a consensus among IPC manufacturers to have a standard wide dynamic function.

Mist penetration

With the deterioration of domestic smog weather in recent years, the market demand for fog-penetrating cameras is very strong. Since optical fog-penetrating lenses are relatively expensive, in order to reduce the price of fog-penetrating cameras and achieve better fog-penetrating effects, mainstream IPC manufacturers have begun to study the camera video image fog-penetrating algorithm technology. Algorithmic fog-penetrating can judge the concentration of haze by the gray-white degree of the local area based on the physical formation model of haze, thereby restoring a clear haze-free image. Algorithmic fog-penetrating can retain the original color of the image and greatly improve the image fog-penetrating effect.

Smart Analysis

HD network cameras started with two or three intelligent analysis functions, such as motion detection and video occlusion, in 2011. Today, almost all mainstream manufacturers’ HD network cameras have more than 10 standard intelligent functions. Of course, most of these intelligent functions are currently limited to mid-to-high-end industry products. According to market business applications, these intelligent analysis functions can be divided into the following points:

1. Diagnostic intelligent analysis. The diagnostic intelligent analysis of high-definition network cameras is mainly aimed at accurately analyzing, judging and alarming common camera failures such as black screen, blur, PTZ out of control, frozen screen, etc., as well as video signal interference such as scene changes, objects left behind/disappeared, etc. Diagnostic intelligent analysis technology is relatively simple to implement. Usually these intelligent functions are integrated in the front end. Of course, the back end such as NVR also has similar diagnostic intelligent functions.

2. Recognition-type intelligent analysis. This technology of high-definition network cameras tends to analyze and process static scenes. Through core technologies such as image recognition, image comparison and pattern matching, it can extract and analyze relevant feature information of people, vehicles, objects, etc. The main application of vehicle recognition and analysis is license plate recognition technology. License plate recognition technology is widely used in parking lot entrances and exits, highway toll stations, etc., and has developed rapidly in recent years: in conjunction with the traffic electronic card system, license plate recognition technology is widely used to capture vehicle traffic violations, effectively reducing the number of vehicle traffic violations and greatly reducing the occurrence of traffic accidents.

3. Behavior-based intelligent analysis. This technology of high-definition network cameras focuses on the analysis and processing of dynamic scenes. Typical functions include: vehicle reversing, zone intrusion detection, personnel focus detection, tripwire crossing detection, fast movement, personnel wandering detection and passenger flow statistics. Mobile motion detection (VMD) is the "early intelligence" in this type of intelligent analysis. VMD makes judgments based on the movement changes of pixel blocks in the video screen. Since it is a two-dimensional image intelligent analysis, the false alarm is high and it is impossible to identify whether the moving pixel blocks are interference or targets. In addition, due to the differences in algorithm technology among security manufacturers, the accuracy of behavior-based and recognition-based intelligent analysis is generally not high.

High resolution

In 2010, the first year of high-definition, only 720P high-definition network cameras were launched. Until 2012, the mainstream security companies' high-definition network cameras were still mainly 1.3 million and 2 million, and there were few with more than 300 pixels. With the introduction of CMOS technology, high-definition network cameras have been boosted. After 2013, 3 million, 4 million, 5 million, 6 million, and 12 million pixel cameras have sprung up like mushrooms after rain. This is the innovation of technology that has brought innovation to the product system. In the existing mainstream security companies in the "Pearl River Delta", "Yangtze River Delta", and "Bohai Rim", high pixels have become the standard configuration of almost all manufacturers, and even 4K has become the standard configuration of all manufacturers. For users, 4K is not just a visual experience and enjoyment, but the resolution of 4K is 4 times that of 1080P. If a 4K camera and a 1080P camera are used to shoot the same scene with the same field of view, the 4K camera will use 4 times the amount of information used by the 1080P camera to restore the scene, and the picture will naturally be clearer and closer to reality. From the perspective of "use", since the amount of information in 4K images is four times that of 1080P, more accurate intelligent analysis can be achieved based on more information. Once 4K is deployed on a large scale, the accuracy of intelligent analysis will be raised to a higher level, and richer and more surprising intelligent applications will be realized.

Starlight

It is generally recognized in the industry that 0.001Lux and below are called starlight cameras. The most representative starlight camera is the TI DM8127/Ambarella S2+Sony IMX185 hardware solution, which is currently widely used in safe cities, finance, hotel buildings, safe villages, ports, highways and other projects. It does not require large-scale installation of fill lighting facilities to obtain better nighttime high-definition color monitoring images. Starlight-level illumination monitoring technology is mainly affected by factors such as lenses, image sensors, and back-end image processing technology. Security companies and manufacturers are also improving from the following aspects:

1. Use a large aperture lens: The lens is an important component of the camera. Its role in low-light monitoring application technology is to focus the light of the target for the camera. The key to the low-light application and technology here is that the larger the aperture of the lens, the more light it will let in. In other words, increasing the lens aperture can effectively increase the amount of light entering, so that the camera can achieve an ideal low-light effect.

2. Choose a large-surface sensor: The essence of a camera is to convert light energy into electrical energy, and the core component of quantification is the sensor. The function of the sensor is to convert light of different intensities transmitted to it into photoelectric conversion, convert it into voltage information and finally generate digital image information. The part of the sensor that receives light is naturally the core of the core. If the camera has the same resolution, the larger the image sensor target area, the greater the amount of light entering per pixel, the stronger the ability to suppress noise, and the better the image quality when shooting in low light.

3. Good image processing technology: In the past, cameras used traditional 2D algorithms to achieve noise reduction, but now 3D noise reduction technology is used. On the basis of the original intra-frame noise reduction, the noise point position is found by comparing and screening the images of the previous and next two frames, and the gain is controlled. The 3D digital noise reduction function can reduce the noise interference of weak signal images. Since the image noise appears randomly, the noise in each frame is different. 3D digital noise reduction automatically filters out non-overlapping information (i.e. noise) by comparing several adjacent frames of images. For cameras using 3D noise reduction, the image noise will be significantly reduced, and the image will be clearer and more thorough, thus showing a relatively pure and delicate picture.