Artificial Intelligent Surveillance – Cross Country Video Monitoring and “CNN based” Face Recognition with Access Control System
Thai Airways International Public Company Limited is the national carrier of the Kingdom of Thailand. It operates domestic, regional and intercontinental flights radiating from its home base in Bangkok to key destinations around the world and within Thailand. From its hub at Suvarnabhumi Airport, THAI (including subsidiaries) flies to 84 destinations in 37 countries, using a fleet of over 90 aircraft. The company's fully paid up share capital amounts to 16,988,765,500 Baht (9 May 2005) and is 53.77% owned by the Ministry of Finance, the Thai Government. At the end of September 2004, consolidated total assets of the company amounted to 193,211 million Baht. In its operations, THAI has achieved profitability every year for the last 40 consecutive years.
ioNetworks INC. founded in 2014 is a professional video solutions provider focusing on high end projects, large scale bids and cloud video services with local telecom companies over APAC and EMEA.
With-in two years, ioNetworks solutions have been conducted in TOP 3 Taiwan System integrators deployed in national banks, intelligent factory, government units and public transportation system. Moreover, ioNetworks has successfully developed channels in Thailand, Malaysia, Vietnam, Dubai, Oman, Jordan, Kuwait focusing on telecoms and valuable projects.
In 2016, ioNetworks officially became INTEL IoT partners with our network video server and video management platform (software) and also awarded with Top10 potential start-up company by the vice president of Taiwan.
We expect ioNetworks video solutions to become world-class leading brand within 5 years providing users more convenient services and more secure life via “AI based video analytics” as well as facial recognition, fire and smoke detection, fall detection for health care, “Cloud based video surveillance platform”, “Video management software”, “Video servers”, and “Central management platform.”1. Major Services and Application: AI Based- Video Analytics Solution
A. EZ Match Face/ EZ Match Fire & Smoke/ EZ Match Face Classification
B. Digital Video Management Solution (EZ Pro Platform)
C. Central Management System (EZ Monitoring)
2. Features of EZ Pro Video Management Platform (SW+Server):
2-1. P2P architecture to easily connect all the devices on the internet and supported “Multi Operating Systems”: (Windows XP/7/10/Embedded, Ubuntu Linux 12.04/14.04/16.04, Mac OSX 10.6 or later (client only)
A. High flexible open platform:
●Support any ONVIF Profile S Device
●Support any RTSP video stream
●Integrate any device with ioEZ SDK, API
B. Robust: EZ Pro has auto failover function controlling limited connected servers to backup for each other without any chance of single point failure. Hardware Server (NVR) is using industrial level components to secure high quality.
C. High performance: Based on the mathematic vector processing framework, EZ Pro could avoid attenuation and enhance the hardware performance to support more cameras recording and display.
D. Expendable: Open SDK and API to easily integrate with other vertical markets as well as smart health care、smart green technology、smart home、access control system、smart retail and so on.
E. Efficient: Installation, setup, and system expansion is simple and takes just minutes with instant automatic camera and server discovery. Revolutionary flexible drag & drop user interface.
F. Connectivity: High compatibility with 90% security surveillance brands as well as Acti, Lilin, Arecontvision, Axis, Brickcom, Canon, Dahua, Digital Watchdog, D-Link, Hikvision, IQinVision, ISD, Sony, Stardot, Vista, Vivotek, and more.
G. Cost saving: EZ Pro failover functions enable any working servers as a failover server to manage all cameras recording. No cost to spare servers and central management server.
H. Enterprise smart search: Use any search feature you prefer - smart motion, keyword(s), time-slice, or calendar to find and share events.
I. Universal Fisheye Dewarping: Automatically or manually detect and dewarp any fisheye lens - including panamorph lenses.
J. Server Health Monitoring: Drag & Drop servers onto flexible grid interface to view CPU, network, memory, and HDD status.
K. One-Click System-Wide Updates: Upgrade your entire system - servers & edge cameras with the click of a button.
2-2. AI Based- Face Recognition Solution (SW+Server):
A. Developed by convolutional neural network, EZ Match could adapt different layers based on different project requirements to achieve high recognition accuracy with more difficult conditions as well as recognizing people with mask or hat.
B. Server(database) and Client (UI/UX) Framework: Convenient remote control.
C. Black List：While suspect appears, system would pop out SMS, EMAIL or Sound alarm via EZ Pro attached a real time snapshot.
D. White List(VIP): While VIP appears, system would pop out SMS, EMAIL or Sound alarm via EZ Pro attached a real time snapshot.
E. Group control: Suitable for access control, door control purpose. Define the employee authorities to each doors or areas. If unauthorized person intruded, system would pop out SMS, EMAIL or Sound alarm via EZ Pro attached a real time snapshot.
Thai Airway has worldwide offices、aircraft hangars and warehouses in global airports. In order to achieve central video management, Thai Airway requires to stream all the camera images back to surveillance center crossing countries. However, by doing so, the storage of recording archive onsite and back to the center and the bandwidth for streaming would double the cost. Thai Airway was seeking for new generation connectivity framework to reach this inquiry with acceptable budges, thus conducted ioNetworks INC. EZ Pro platform to link all the IP cameras from worldwide offices.
Link all the cameras from worldwide offices by IP address1. After Connecting all the cameras：
A. All the servers、IP cameras、archive (storage) form a big system.
B. After log-in by different authorized account, administrator could search、watch live view、monitor, export archive and do any actions, in contrast users could only do the actions that administrator assigned.
C. Each server would check the status of each other every 30 seconds. If a server down, another working one in the system would do the failover by taking over the cameras recording from failed server within 45 seconds automatically.
D. Without streaming back to the center for recording backup, EZ Pro save a lot of cost to bandwidth、storage、spare server、24hours maintenance fee.
E. EZ Pro video management software is lifetime usage, plus industrial servers, the whole system is able to run five to seven years without failure. ioNetworks only charges the service fee and upgrade fee if new requirement needed to aid Thai Airway to save 30% maintenance fee、60% bandwidth fee and 50% new construction fee than before.
A. Face Recognition Services – access control and suspect alarm to worldwide offices.
B. Implement face recognition after punching in, to avoid absent and cheats.
C. Recognize 8 faces within one second. Plan to replace punching behavior in the future.
D. Enroll suspects or recidivists image data into the EZ Match Face system for alarm purpose.
1.Difficulty to integrate with original system: Exiting platform and IP cameras are out of day. With flaw firmware, the streaming of exiting cameras is incompatible to latest ONVIF standard and failed to perform perfectly in EZ Pro. Besides, exiting CCTV software is a close system, without open API or SDK, it is literally not working with ioNetworks video management platform.
A. Work with camera provider to maintain the latest firmware with standard ONVIF profile S which is 100% compatible with ioNetworks video management system.
B. Since the exiting platform is a close system, we set up a three-years license leasing plan to replace all the exiting devices with affordable price at the first beginning.
2.High difficulty to sample face dataset: Due to the high height between cameras and the ground、low lux onsite and vague facial features localization of 1000 people database, the recognition accuracy failed to reach 90%. In order to achieve 95% accuracy, the sampling dataset to each person requires a two-minutes video with 24 convolutional neural network layers (features filter) that prolong deep learning process and make convergence more difficult.
A. Form a sampling SOP to aid clients accomplishing effective face dataset within one week.
B. Purchase Nvidia Tesla P100 high end graphic card to shorten the training process.
- Initial Kick-off, Planning Meeting
- Discussion with Thai Airway Planner and Aviation Police
- Images from surveillance IP cameras to each different spots
PS: Due to the confidentiality, the content of face recognition service is not able be revealed.