HIKVISION AI Products Introduction: Facial Recognition Applications
Technical Development of the Hikvision Research Institute (HRI) Deep learning Deep learning AI technology begins to Basic intelligence and The algorithm team is contributes to and initiated, AI accelerate in many industries industry intelligence formed and the intelligent enables the security research begins to solve industrial pain points algorithms released algorithm research begins industry 2007- 2012 2013 2014-2015 2006 2016-2020 License Passenger flow License Clustered training Video Open AI platforms number statistics number platforms structuring recognition recognition Behavior analysis Traffic incident and facial DarkFighterX Binocular Facial detection recognition cameras Audio recognition camera recognition based on DL ATM protection Self-driving vehicles Smart265 Natural language Perimeter …… processing protection …… Multi-dimensional big data …… …… ……
Three Key Factors for Artificial Intelligence Basic Data The volume and diversity of data are growing rapidly. Hikvision has accumulated abundant CCTV-based data. Computing Ability Data Computing Computing abilities are growing along with the diversification of AI chips. Collaboration with the world’s first-class GPU vendor. Algorithms Algorithms Deep learning research continuously drives algorithm optimization. More than 1,000 engineers have now been working on AI algorithms for several years.
AI Cloud Platform – First CCTV Company to Launch Its AI CLOUD in 2017 Edge Nodes Edge Nodes focus on multi-dimensional data perception acquisition and front-end intelligent processing. Edge Domain Edge Domain focuses on data convergence, storage, processing, and intelligent applications. Cloud Center Cloud Center focuses on the convergence of multidimensional data, including IoT data, and multidimensional analytics applications based on big data.
Hikvision AI Training Platform – Enabling Personal AI Applications Workflow: Continually increasing imaging in actual use, forming a closed data loop Built-in algorithm network, semi-auto Collect and annotate training images Import Generate algorithm Algorithm warehouse NVRs IP Cameras User Images Training Implementation Applications: Intelligent Banks Mines Healthcare Military Manufacturing
Abundant AI Products AI Products Front End Back End Servers VMS p Intelligent PTZs Big Data Servers HikCentral Master Behavior Analysis NVRs Fusion Intelligent 8 IPCs Full Analysis Servers HikCentral 7 Intelligent IPCs Enterprise DeepinMind NVRs Intelligent Fusion Servers 6 Special IPCs Behavior Analysis HikCentral Servers Professional 3 AcuSense IPCs
Attribute Face Face Face Face Face Result Face Extraction, Video Detection Tracking Grading Capture Comparison Output Output Face Modeling Face Capture Facial Recognition
Up to faces that appear in the screen can be detected and captured.
List alarm Stranger alarm High frequency alarm Face attendance 1 V 1 comparison Identity verification Search by face picture Face attribute analysis and search by face attribute Personal archive Privacy protection
• Recognizing faces in real time and giving alarms on people in face libraries, such as VIPs and blacklists • Blacklist alarm: preventing potential criminals • VIP detection: recognizing potential customers to help boost the sale Alarm Linkage Captured face Compared with blacklist/VIP list • Scenarios: retail stores, banks, stations, office Matched! Alarm! buildings, hotels, etc. Retail Bank Blacklist Alarm VIP Detection Station Office Building
• Recognizing faces in real time, gives alarms on strangers (people that are not in the face library) • Keeping strangers away from open but sensitive areas. • Scenarios: important places where only insiders are allowed to enter, such as office buildings, Alarm Linkage Captured face Compared with blacklist/VIP list Not Matched! Alarm! residences, laboratories, warehouses, vaults, etc. Vault Warehouse Office Building Laboratory
• Compares captured faces and triggers alarm for frequently appeared people 1 • Public security scenario: finds potential criminal, 2 pre-alert 3 … • Commercial scenario: finds potential consumers Alarm Linkage Compared with all captured face pictures Reach the set to increase the sale Captured face Count appeared times frequency! Alarm! • Scenarios: retail, 4S shops, luxury stores Retail 4S Shop
• Realizes non-sensing attendance check efficiently • Able to filter internal staff • Able to associate with multiple channels and face libraries Matched, have Captured face Compared with face library • Scenarios: entrances of office buildings, attendance records meeting rooms, classrooms, etc. Attendance Report Friendly UI Office Building Classroom Meeting room
• Choosing two pictures (captured or uploaded picture), comparing them, and getting the similarity 97.67% • Quickly confirming if it is the same person • Scenarios: airport customs, subway stations, police similarity stations, etc. captured picture passport picture Airport customs
• Choosing one picture (captured or uploaded picture), comparing it with all the pictures in the face library • Usually compared with face library of criminals or population database • Quickly confirm the identity of someone Choose one Get the picture with highest Compared with face library • Scenarios: airport customs, subway stations, police similarity, picture Confirm identification stations, etc. Station Main road
• Searching by a face picture, and getting all the similar captured pictures with similarity, time, location, etc. • Able to generate the face track on a map according to the search results (server) Uploading face Finding all the similar pictures captured pictures • Quickly finds a target • Scenarios: shopping malls, stations, residences, business parks, key streets, etc. Shopping mall Residence Business park Station
• Searching by attributes of faces, and getting all the Gender matching captured pictures. Age Glasses • Face attributes (optional): gender, age, glasses, Smiling smiling, wearing mask Wearing mask Finding all the • Quickly finds a target captured pictures that Choose attributes meet these attributes Shopping mall Residence Business park Station
• Creating files for everyone and putting all the captured face pictures of one person in one file, which contains time and Facial Recognition and space information of the person. clustering • Facilitating people management Real Name Jack Jack, 2019.1.1 00:08 , Camera A Jack, 2019.1.1 00:09 , Camera B Jack, 2019.1.1 00:13 , Camera C Not Real Name Pedestrian A Pedestrian A, 2019.1.1 00:08 , Camera A Pedestrian A, 2019.1.1 00:09 , Camera F Not Real Name Pedestrian B Pedestrian B, 2019.1.1 00:15 , Camera B Pedestrian A, 2019.1.1 00:18 , Camera F
• Human face/body region info. and marker info. Video • Realize the function of stream mosaic live view through Live View inherent play library DeepinMind NVR HikCentral • Select mosaic targets (Customization) (Customization) • Drawing mosaic on • Enable mosaic channels face/body area • Enable fixed area mosaic • Overlay coordinate area info. Export Video
Face Capture 3 Series AcuSense DeepinView 7 DeepinView 8 DF Series PTZ 8 Series – Fusion iDS-2PT9A144MXS-D/T2 Cameras Series Series Cameras Intelligent Face Recognition iDS-2PT9A144MXS- DeepinMind NVR Intelligent Fusion Full Analysis Server DeepinView 7 DeepinView 8 8 Series – Fusion D/T2 Server Series Series Intelligent
Highlights: iVMS 4200 Baseline • Device management Client • Easy to deploy • Real-time alarm • Cost-effective Switch Functions • Face capture Facial Recognition Camera • List alarm • Face attribute analysis • Real time face capture • Face comparison
7 Series 8 Series 8 Series – Fusion Intelligent iDS-2PT9A144MXS-D/T2 • DarkFighter & LightFighter • DarkFighter & LightFighter • Supports multiple intelligent • Supports multiple intelligent • Multiple network security • Multiple network security functions at the same time functions at the same time methods methods • Integrates multiple lenses to • PanoVu PT camera is able to • Switchable to support • 5 defined streams + 5 custom see the details while get both panoramic views and multiple intelligent functions streams monitoring the large view close-ups with the integration • 5 defined streams + 5 custom • Supports providing metadata design of a PanoVu camera streams based on ONVIF & ISAPI and a PTZ camera • Supports providing metadata based on ONVIF & ISAPI
Recommend
More recommend