Interview camera module manufacturers: face recognition in the single camera and dual camera how to select

With the increasing popularity of intelligent hardware, its product types are also more diversified to adapt to more demand scenarios. Taking the camera, an important part of face recognition, as an example, various products are not only different in price and style, but also have monocular and binocular distinctions. So how do you choose?

Recently, in the 2020 High-Tech Fair, the author introduced He Shikun, general manager of camera module manufacturer Jinshi Kangkang, which focuses on the image acquisition industry and has strong professional ability, to conduct a deep communication on how to select a face recognition camera.

Key point 1: Choose single or binocular? Specific depends on the need for live detection

Monocular camera is mainly RGB camera, with monocular algorithm can quickly complete the acquisition, the high-quality image is sent to the back-end for identification and comparison, suitable for face defense, accurate promotion, passenger flow statistics, access control management and other application scenarios. This kind of camera can achieve monocular live detection with the appropriate living detection algorithm. For example, by introducing the ArcFace face recognition algorithm free and open on the Hongsoft Vision open platform, Jinshi Kang realizes silent live recognition without user action cooperation. In addition, with the help of ArcFace algorithm, it can also quickly realize a full set of functions such as face recognition, age detection, gender detection, and face recognition under large area occlusion.

The binocular camera is based on the RGB camera, adding an IR infrared lens, both near infrared light and visible light video acquisition function. Based on the principle of infrared imaging, the screen class cannot be imaged, so it has a natural ability to resist the fake face attack of screen imaging.

Therefore, compared with the monocular camera, the binocular camera is better in the defense ability against the fake face attack. For example, the 5 million binocular wide dynamic face recognition camera sold by Golden Vision Kang uses the IR binocular living technology of Hongsoft free ArcFace algorithm. Based on the infrared imaging principle and the high robustness of deep learning algorithm, it has excellent defense against paper photos and screen photos.

However, due to the differences in hardware, the cost of binocular living objects is relatively higher than that of monocular RGB living objects. In general, the monocular face recognition solution is easy to deploy and low cost, while the binocular face recognition solution is more accurate in live detection, and is more adaptable to factors such as light changes and complex background environment, which can support true unattended. Therefore, in the selection process, specific considerations should be made according to the actual application scenario, the live effect you want to achieve, and the final cost, to balance the needs of all parties, and cannot be generalized.

Key point 2: Camera imaging core components and key parameters

Among them, the resolution power refers to the clarity that the camera module can achieve and the detail expression ability of the image. A more intuitive evaluation method is to visually identify the black and white line pairs in the figure by shooting the ISO chart test, and human eyes can identify the limit line values of the black and white line pairs. In the same environment, it is also necessary to check the image for abnormalities such as flower screen, lines, and color bias. In the whiteboard test environment, the dark Angle is usually checked to detect the brightness uniformity of the module, detect whether there is dirt abnormal in the lens, and detect whether the chip has bright spots and bad spots abnormal.

In all these links, testing is not only a Bug, but also an important step in the development and improvement of product design, so the testing work is often highly professional and rigorous to the product effect. If you want to save this large number of selection, development, testing cycle, or suggest that the demand side directly to the Hongsoft vision open platform industrial chain market for direct procurement.

He Shikun, general manager of Jinshikang, concluded: "In recent years, we have developed reasonably well in the field of faces, with an annual growth rate of more than 50% from 2018 to 2020. Through cooperation with Hongsoft, we have also made a lot of different programs, different costs of camera modules, mainly to cope with different choices of different customers, to facilitate their simple and low-cost use, but also to reduce the threshold for more customers who want to do artificial intelligence field to enter the industry."

The working principle of the camera is to convert the light signal into an image signal, so the complex light environment such as strong light, backlight, and dark light will directly affect the image quality of the camera. The quality of the camera imaging will also directly affect the effect and experience of face recognition and living detection. Therefore, when making product selection, developers should also pay attention to the core components and key parameters of the camera imaging to ensure that the quality of the image collected by the front-end meets the requirements of face recognition.

In the three core components of the face recognition camera module, lens, image sensor and DSP image processor, the lens mainly assumes the task of capturing the object and focusing the shot image on the image sensor, and its main parameters are field of view Angle, aperture, CRA, distortion, glare and so on. The face recognition camera module generally adopts a fixed focus lens with fast focusing speed, stable imaging quality and accurate metering.

For the image sensor which is responsible for converting optical signal into digital signal output, its main parameters include sensor size, effective pixel, pixel element size, etc., CMOS sensor with high integration, low power consumption, fast speed and low cost is recommended. Finally, the digital image signal is optimized and transmitted to the face recognition device DSP image processor, high-quality DSP can improve the original chip wide dynamic range by more than 30%.

In order to further facilitate developers, Golden Vision Kang launched the integration of dynamic scene, backlight scene, highlight scene and as many as 6 scenes of the "multi-scene mode", the multiple scene modes are compatible with a camera, through the camera automatically judge the light environment in the scene, and then call the corresponding mode, in order to achieve the best face recognition effect in a particular scene.

According to reports, this is also one of the most popular products in the face recognition camera module launched in the Hongsoft vision open platform industry chain market. The industrial chain market is a commercial platform launched by Hongsoft Vision open platform this year to focus on building industrial upstream and downstream ecology, and has now launched a series of high-quality products from algorithms to hardware, from whole machine products to various industry solutions, which can achieve accurate docking for the supply side and the demand side, and effectively shorten the time of face recognition products from demand to landing.

Key point 3: How to test the basic functionality and reliability of the product?

To ensure that the product can be used smoothly after landing, testing is an essential part of it. As far as the camera module is concerned, its basic performance test includes the analytical force test in the normal shooting test environment, the screen test, the line test, the color test and the dark corner test in the whiteboard test environment, the dirt test and the bad spot test.

Camera Module Solution Service Provider

Tel:+86 186 6497 2870

WhatsApp:+86 186 6497 2870

E-mail:caida@dgzx.hk

https://dgzx.hk