Artificial intelligence (AI) is the core topic of the Expo 2018, of which face recognition-related products and solutions are still the most displayed part of AI in the field of security monitoring applications at the Expo. At present, video surveillance has entered the application stage of face recognition and vehicle recognition. Everyone is concerned about the improvement of the accuracy of the recognition algorithm, and how to improve this accuracy, aside from the level of the algorithm, from the perspective of the hardware structure of the equipment, the first thing that needs to be guaranteed is Monitor image sensor performance.
In this Expo, the author also had the honor to interview Mr. Bai Zhendong, general manager of the artificial intelligence division of SmartSens, a domestic enterprise focusing on image sensor technology and product research and development. During the interview, Bai Zhendong analyzed the current domestic image sensors in detail. The innovative technology highlights and application status have given us an additional layer of cognition and understanding of the core components inside this security video surveillance camera.
Face recognition applications put higher demands on image sensors
Due to the particularity of the industry, the image sensor of the security video surveillance camera mainly considers three key points-signal-to-noise ratio, sensitivity and dynamic range. At this stage, applications such as face capture and video structuring emphasized by intelligent security monitoring have put forward higher requirements for the "vision" system.
"The sensitivity of the image sensors used in the traditional security monitoring field is difficult to meet the complex application scenarios of today's intelligent security, especially the requirements for accurate capture and recognition of face images. The traditional security monitoring cameras using such sensors are in order to obtain a more ideal Imaging brightness often increases the shutter time used for shooting to more than 40ms, and uses a relatively high gain to ensure imaging brightness. However, during the shooting process of the camera, people are constantly moving, so a long time The shutter time will cause blurred face images, and higher gain will lead to image quality degradation and increased noise, these two factors seriously affect the output quality of face images and even lead to poor final recognition results." Bai Zhendong said.
Just as mobile phones are moving from "functional" to "smart", the security field is currently undergoing such an upgrade.
"In view of this industry development trend and the specific application form of intelligent monitoring that accurately captures/effectively outputs feature information of faces, vehicles and objects in monitoring scenarios, we are currently launching the '1/1.8' series of super-sized star-level images. Sensor products, this series is a subversive product designed for intelligent security, which uses the BSI back-illuminated process that can improve the sensitivity of CMOS image sensors, and adopts a unique new sensor design architecture in the design, which further improves the image. The sensitivity and signal-to-noise ratio of the sensor enhance the ability of dynamic range output, which fully guarantees the adaptability of smart security cameras to complex scenes (dark light/backlight)." Bai Zhendong added.
Another technique that improves image sensor performance can be achieved by changing the way the imaging exposure is made.
Two exposure Shutter technologies: Rolling Shutter and Global Shutter
Rolling Shutter: The exposure method of Rolling Shutter is to expose in a row-by-row manner, that is, when the sensor is imaging, each row of pixels is exposed row by row, and there will be a certain time interval between the exposure of each row of pixels.
Global Shutter: Unlike Rolling Shutter, the whole picture is exposed at the same time. All pixels of the image sensor in the camera sense light at the same time and expose at the same time. The sensor reads the same picture at the same time.
High-speed object deformation
(The picture on the left is a picture taken by the SmartSens SC132GS with Global Shutter technology, and the picture on the right is a picture taken by the Rolling Shutter sensor).
chiaroscuro
(On the left is the shooting effect of SmartSens Global Shutter CIS with single-frame HDR technology)
According to Bai Zhendong, Rolling Shutter is commonly used in traditional cameras, which is a progressive exposure method. For example, an image sensor is a front, and its exposure number is row by row exposure, which is suitable for static or slow flow scenes. However, in dynamic scenes, especially in the face of high-speed moving target objects, the Global Shutter exposure method needs to be used at this time. Its exposure is that the entire picture is frozen in one frame at the same time, no matter how fast the picture speed is, as long as the shutter setting is reasonable , can be photographed very clearly.
Global Shutter technology that best matches AI vision applications
It is understood that the Global Shutter image sensor technology is the earliest high-end technology applied in the field of industrial automation. It mainly captures high-speed and clear images for automated industrial production line inspection systems, and provides accurate image data sources for moving robotic arms. With the rise of the AI ​​era, more and more application fields need to use intelligent perception products with edge computing capabilities.
"As for the market development trend of artificial intelligence, we have already begun to predict around 2015. If artificial intelligence is used to make some revolutions in this industry, the application of Global Shutter technology is an inevitable trend. The Global Shutter actually has fill light on the scene. Its environment is relatively stable, but the artificial intelligence scene is uncertain and the environment is more complex. Therefore, on the basis of Global Shutter, we also added BSI back-illuminated Sensor to improve the sensitivity of the sensor. Putting BSI technology and Global Shutter technology together can greatly enhance the scene applicability of Global Shutter." Bai Zhendong said.
At present, the technical combination of Global Shutter+BSI has been applied in multiple vertical application fields, assisting the application of smart vision in artificial intelligence scenarios, such as smart transportation, smart business, smart perception and other fields, with practical applications in the field of smart transportation For example:
Smart Transportation
From a macro perspective, in order to make urban transportation more convenient and ease the increasingly congested public transportation, smart transportation system is one of the effective means to solve the bottleneck of transportation development. Imaging technology is inseparable from application scenarios such as high-speed violation investigation and punishment, ETC at bayonet, road monitoring at intersections, intelligent management of parking lots, road flow monitoring, and electronic police. Only with accurate imaging information can the current road conditions, vehicle behavior and vehicle specific identity information be judged more accurately, and the smooth operation of the urban road traffic system can be better achieved. In this area, the Global Shutter CMOS image sensor can play a better role.
From a micro perspective, automotive products are also evolving from being equipped with ADAS-assisted driving to autonomous driving. Whether it is assisted driving or automatic driving, the central control chip of the car has higher and higher requirements for the image information received by the sensor. The central control chip not only needs the image sensor to provide clear pictures without deformation in the high-speed motion of the vehicle, but also needs to provide the distance between the surrounding objects and the vehicle, and even monitor the driver's fatigue level. The Global Shutter sensor can avoid the problem of image deformation when the vehicle is moving at high speed. At the same time, it combines the algorithms and logical judgments of intelligent driving/autonomous driving/driver fatigue detection to provide a great guarantee for driving safety.
Value Evolution of Image Sensors: From Enablement to Service
Although the current first year of artificial intelligence has begun, in Bai Zhendong's view, it is still in the stage of weak artificial intelligence at this stage, and the intelligent vision system relying on image sensors still has a long way to go.
"In the era of weak artificial intelligence, people use sensors more for intelligent perception of environmental information. After the perception system is perfected, they will gradually expand deeper intelligent applications in vertical application scenarios." Bai Zhendong said.
Judging from the application of sensors in the field of artificial intelligence and machine vision at this stage, the main functions of sensors are imaging for people to see, and imaging for machines. Face payment, face unlock and other scene applications. From a set of visual perception tools to the role of serving the daily life of the public, the application value is also continuously evolving with the continuous enrichment of scenes, and these are exactly where image sensor companies make their efforts.
"The fields of artificial intelligence and machine vision involve the application of image sensors in a wide range, and the user's requirements for image sensors in terms of cost, performance or application cooperation will also become our continuous driving force in the process of new product development. , through our unremitting efforts in the field of image sensor technology, to better serve the development trend of application technology and create a better life for the public!" Bai Zhendong finally emphasized!
Shenzhen Xcool Vapor Technology Co.,Ltd , http://www.xcoolvapor.com