The hottest Yixun AI intelligent video double reco

2022-10-22
  • Detail

With the rapid construction of 5g network and the rapid development of AI intelligent technology and video communication technology, the integration of various technologies has brought new solutions, services and more financial business scenarios. With the help of small program video platform, this scheme realizes seamless integration with AI intelligent dual recording platform, realizes AI virtual robot intelligent dual recording system, improves user experience and reduces labor costs

1.2 scheme overview

this scheme adopts the seamless integration of AI intelligent technology, virtual human technology and video communication technology to provide AI intelligent virtual human double recording solution and serve car loan users through 4g/5g communication link

1) ASR speech recognition technology: recognize the questions answered by users through ASR

2) TTS text to speech technology: broadcast business process scripts through TTS

3) business process configuration management: business processes can be dynamically configured, and parameters of scripts in related businesses can be customized

4) video customer service communication platform: it provides real-time two-way video interactive service capabilities. The platform has the functions of operational amplifier destruction, virtual human video technology, and real-time double recording

5) virtual human and video customer service platform integration technology: virtual human and users conduct video technology integration, virtual human voice synchronization, "providing a real user experience

6) applet access: provide the communication access ability between applet video and video customer service platform

7) face recognition technology: recognize the user's face in real time to ensure that it is handled separately for the user in the process of business handling

8) applet SDK: provide high-level sdkapi, which is easy to use and can be quickly integrated with applets

9) business process management: intelligent business process management, according to the business scenario

1.3 technical architecture diagram

2 Demand analysis

2.1 Business scenario description

users handle the video audit and recording of car loan business through a small program. After connecting with AI video customer service, they verify their identity through face recognition technology. Through the integration of TTS technology and ASR technology, they answer the questions in the audit process until the business is completed. The whole process of business processing shall be recorded and recorded, and the business processing data shall be retained and submitted to the business system for manual or intelligent review

2.2. Key function

face recognition, when the user's face is inconsistent with the face of the order handling user, it will prompt the user not himself (photo 1:1 comparison, the comparison source is provided by the applet)

face confirmation, When the user's face is not in the picture, it prompts the user to keep the face in the center of the picture (photo 1:1 comparison function, to ensure that the face is in the picture). The above function points have the risk of small programs on the shelves (see 4.1)

video calls and video recordings support logo watermark and time watermark. (see 4.2.1)

Video supports 480p definition (see 4.2.1)

video recording, supports the recording of virtual human images and user images (see 4.2.2)

supports the replacement of virtual human image background (see 4.2.3)

video quality inspection: capture faces every 30s and send them to face recognition suppliers for comparison, If it is inconsistent, it will prompt to keep the face consistent (see 4.2.4)

support the acquisition of location and display it on the page (see 4.2.5)

support pop-up notification after interruption, and select exit or automatic exit after pop-up (see 4.2.6 applet optimization exception handling module)

after the video recording is completed, place it in the specified location, Support downloading and calling (see 4.3.1 order management)

support script configuration and intra script support variable parameters (see 4.3.3)

the customer answers the positive and negative specific parameter values of semantic conversion (the face-to-face configuration management module needs to add the voice semantic conversion module)

if the recording of the same order fails for the first time, fails for the second time, and succeeds for the third time, the first and second recorded files are saved for X days, and call view is supported, Order management supports that one order corresponds to multiple video problems

3. Architecture design of remote video face-to-face signing system

3.1 System design principle

stability principle: the product is mature and stable, and the system should have sound safety precautions. For key applications or modules to achieve equipment redundancy, application clusters, database clusters, the system should be able to operate stably in normal and peak business processing, and provide continuous and reliable services. Realize flow control, error handling, retransmission mechanism and timeout control to meet the requirements of real-time online transactions under large data volume and large transaction volume. Batch data processing and online transaction processing do not affect each other, ensure the normal operation of the system for 7 * 24 hours, and ensure the continuity and stability of the system operation

security principle: the system should adopt a perfect security and confidentiality mechanism to ensure that all kinds of data are not destroyed, illegally accessed and maliciously modified, and to ensure the correctness, integrity, consistency and security of customer and bank transaction data

practicality principle: the system must ensure its practicality, practically meet the construction requirements of the bank and fully meet the business needs. The system shall have a friendly user interface and be easy and efficient to operate

maintainability principle: the system design should be easy to maintain, follow the principles of modularization, componentization and parameterization, be easy to transform and expand, be flexible to set, and be convenient for maintenance and management. The system can provide a comprehensive system management and maintenance platform for technicians to maintain

transaction monitoring principle: in order to facilitate technical maintenance personnel to master the operation of system transactions. The monitoring should be clear, provide the storage and query of auditable data such as operation log and transaction log, facilitate the system maintenance of technical maintenance personnel, and provide a complete and automatic log backup and cleaning mechanism

data standardization principle: the system construction should comply with the industrial standards of data management, the norms of the financial industry, and the people's Bank of China and the China Banking and Insurance Regulatory Commission The automatic control of the experimental process will meet the requirements of various regulatory standards

openness principle: the software and hardware platform and database system should be open, and the system provides rich external interfaces

efficiency principle: the construction of the system should be forward-looking in business and technology, with reasonable system design architecture, high processing efficiency, low resource occupancy, and avoid excessive data redundancy. It can not only meet the current business needs, but also fully consider meeting the business development needs in the coming years, and has a certain innovation leading ability

scalability principle: the system design should have good scalability, including the expansion of processing capacity and data access. The system development should follow the design principles of componentization, modularization, parameterization, high cohesion and loose coupling, so as to ensure that the software system architecture is easy to transform and expand, and improve the reusability, maintainability and development efficiency of the software, To meet the requirements of the bank for system performance or function improvement in the future development

forward looking for new business: the functions provided by the AI applet video virtual face-to-face signature solution can fully meet the existing needs, without adding redundant functions unrelated to the needs. At the same time, it also ensures that in the next few years, with the innovation of the business, the system can adapt to the changes of needs, and only need to expand the corresponding functional modules or concurrent licenses to meet the upgrading of future business

it can be upgraded to an artificial video customer service platform to provide AI intelligent double recording and artificial video customer service capabilities for the bank. In view of the principle of reusability, the system should have an existing service environment, and the expansion of new service components can meet future business needs

in the future, it can realize unified routing, unified reporting, unified queuing and other functions with the voice platform, so as to save the construction cost in the bank

3.1.1. Applet

3.2 AI video face-to-face signing system network architecture

system network architecture diagram

3.3 Remote video face-to-face sign system performance

3.3.1 System concurrency

after the system is put into use, the first phase supports the access of 10 concurrent users. The system should ensure the stable operation of the system under the condition of supporting 10 concurrent users, and can not saturate congestion; The system design capability supports 32 user single machine access capability

3.3.2. Network performance requirements

1) it has intelligent packet loss resistance and short network delay, so as to ensure that customers can handle business smoothly under the network of Unicom, China Mobile, telecommunications and other operators, and the picture is not stuck

2) under normal network environment, AI video start-up time is 3 seconds

3) the small program network delay is less than 1000ms to ensure the smoothness of business processing

4) the connection success rate is greater than 95%

5) video definition supports 360p-720p

6) meet the customer's ability to launch interactive videos in major regions at home and abroad, with clear and smooth video images without jamming

7) the response time of all statistical analysis functions of the system shall not exceed 3 seconds

4. Main functions of AI video face-to-face signing applet SDK

4.1 Face recognition

users collect face photos in real time during video face-to-face signing, call the face recognition interface to dynamically compare ID photos and face photos, and confirm whether they use their ID cards to handle business for the user

4.2. AI video face-to-face signing applet SDK

the capabilities of the video face-to-face signing applet SDK are as follows:

1) initiate video call: connect AI virtual video customer service and handle face-to-face signing business

2) end video call: end AI virtual video customer service dialogue, and complete or terminate business processing

3) display position: the position information obtained dynamically in real time, provide accurate position information, and upload it to the business system

4) capture face photos during video calls

4.2.1 AI video signature intelligent Q & a

video support 480p definition

virtual human video quality support 480p and above, and the video resolution can be automatically adjusted according to network fluctuations

support video synthesis with watermark. When users handle business, the video displays logo watermark and timestamp watermark

video watermarking support: video files after video synthesis support prefabricated watermarking and video synthesis, and support timestamp watermarking and geographic location information. The video shall contain text or image watermarks, which can be configured and managed on the server

4.2.2. Video recording

the whole process of user handling business is recorded and recorded, and a single MP4 video file is generated

1) user video and audio

2) AI virtual human video and audio

3) user applet UI: user location information

4) user applet UI: timestamp information in the user

5) user applet UI: enterprise logo and background

6) support virtual portrait background replacement

4.2.3 Support virtual portrait background replacement

virtual portrait background supports virtual background technology, can customize background pictures, and can be configured to replace background pictures on the server side

4.2.4 Video quality inspection

business scenario: in order to ensure that the user's face is always in the video window during business recording, the user's portrait can be recorded throughout the process and recorded in the video. Face recognition is needed to judge whether the user's face is in the video window. If the user's face leaves the video window, a prompt will be given

video dynamic face quality inspection: capture the face every 30s and send it to the face recognition supplier for comparison. If it is inconsistent, the user will be prompted to keep it

4.2.5. The location information display

applet obtains the current location information of the user in real time and displays it on the video call page. The location information is uploaded to the business system for recording

4.2.6. Other functions

support pop-up notification after interruption. The pop-up window selects to exit or automatically exits

in the process of business processing, if there is an interruption (the user's reply does not conform to the business process), the pop-up window will pop up

Copyright © 2011 JIN SHI