In recent years, due to the development of deep learning algorithm and the rapid increase of face data, face recognition technology has been widely used. PC, mobile phone, smart camera, security camera, traffic camera and other “machine eyes” are growing at an alarming rate. However, due to the sensor capacity, the 3D information of human face can not be captured, resulting in the traditional 2D face recognition equipment and algorithms can not achieve the recognition accuracy and security of payment level, and are vulnerable to attacks by photos, videos or silicone masks. In addition, the accuracy of 2D face recognition is also greatly reduced in high light and dark light scenes, and the application scenarios are still highly limited.

3D face recognition technology is to collect 3D face data through 3D camera, capture accurate 3D coordinate information of each point on the face in the field of view, and further realize 3D face detection, key point positioning and 3D face recognition through algorithm. High precision 3D recognition technology greatly improves the security of face recognition and widens the application scenarios of traditional face recognition. It can not only be applied to the occasions with high security requirements such as door lock and access control, mobile phone unlocking, financial payment, but also the corresponding three-dimensional processing technology can be expanded to the fields of security monitoring, gesture human-computer interaction, somatosensory interactive entertainment, virtual / augmented reality, three-dimensional display of goods, robot man, automatic driving, etc.

Major companies have been in 3D visual layout for a long time, but there has been no large-scale application. Hardware is the limiting factor, and the complexity of algorithm is the main reason. In order to adapt to 3D depth camera modules from multiple manufacturers and greatly reduce the threshold of 3D application development, Tuyu technology recently launched the 3D face recognition cloud platform. It is reported that the platform currently provides a unified API interface and online tools for free. It only needs to upload point cloud data to use cloud face recognition, point cloud browsing, accuracy evaluation and other functions. It is worth mentioning that at present, many 3D recognition algorithms require to obtain face coordinates through 2D images. The 3D face recognition algorithm launched by Tuyu technology is completely based on point cloud information and does not need 2D data support.

Figure 1- 3D face comparison online interface

The related technologies involved in the platform have their own intellectual property rights. The core algorithm uses the latest point cloud neural network to realize a series of operations such as face detection, key point positioning, face alignment, and solves the scarcity of 3D face data by making full use of 2D image data through migration learning. At present, it can tolerate a depth error of about 2mm. Considering the limited computing power of the embedded terminal and the module acquisition capacity, the graphical 3D face recognition algorithm can also use the hash algorithm to reduce the computational complexity to 1/10. At the same time, the multi frame fusion method can be used to further improve the recognition accuracy.

The launch of Tuyu technology’s 3D face recognition cloud platform has opened the 3D platform cloud algorithm recognition service, enabling small and medium-sized enterprises and institutions in security access control, smart finance, smart home and other industries, greatly improving industry efficiency and building a closed-loop ecosystem. It is understood that Tuyu technology has a professional team engaged in the development of 3D face recognition technology earlier in China. It is committed to technological innovation in 3D measurement, 3D face / object / scene recognition and other visual fields. It provides advanced 3D visual intelligent processing solutions for customers such as security, robotics, vr/ar, payment and industrial detection, and can provide a complete industrial chain of “deep modules + core algorithms + chip acceleration”.

The 3D face recognition cloud platform was launched to help the development of 3D face recognition applications

Figure 2- composition of 3D face recognition cloud frame in figure language

At this stage, graphics technology takes 3dface algorithm and module as the core to develop solutions for door lock access control, mobile phone unlocking and retail payment scenarios. In the future, Tuyu technology will continue to rely on core software and hardware to expand scenarios and applications, promote the market penetration of 3D technology, and explore applications in vehicle, robot, entertainment and other industries. Independently develop high-precision, low-cost and customized software and hardware integrated depth modules and module arrays for near, medium and long-distance and challenging scenes, and strengthen the accumulation of core technologies such as 3D rendering, reconstruction and positioning.

Leave a Reply

Your email address will not be published. Required fields are marked *