The Deep Vision AI platform provides intelligent video analytics using AI through vehicle recognition, people counting and demographics, facial recognition, and others. The following sections describe the Deep Vision AI components:

Admin Console

The Deep Vision AI Admin Console provides a web interface that is used for the administration and configuration of the software, managing cameras, and managing users. It allows the user to add camera streams and enable any of the AI-based software modules that Deep Vision AI supports.

AI Modules

Deep Vision AI has a modular architecture, in which each software module provides an AI capability to perform advanced analytics. Each video stream from a camera can be configured with one or more of the following modules: 

• Facial Recognition module—This module monitors designated areas for a watch list of unique individuals. The images of these individuals are uploaded to the module, after which these individuals can be identified. The designated areas are monitored, and alerts are raised when the individuals appear on video stream. Retail stores can use this feature for notification of VIP customers; corporations can use this feature to track employees’ locations.
• Vehicle Identification module—This module identifies the year, make, and model, of vehicles entering designated areas. Retail stores can use this information to offer better and faster curbside pickup service. Retail stores can map the vehicle information to customers using historical information. The module also records arrival time, departure time, and dwell time for analytics. When applied to cities, this module allows for real-time alerts based on matches with specific vehicle watch lists.
• People Counting and Demographics module—This module records the number of people in a designated area, their demographic information (gender and age), and their dwell time. The retail store can use this information, augmented with other demographic information such as geography, interest, preference, and income, to understand customer segments, influence product selection, and provide product placement, customer communication, and advertisement.
• Objects Entering Alert Zones module—This module counts the number of people or vehicles by type entering a designated area. It provides alerts based on the number of objects and the time spent in an area. The module makes it possible to:

• Count the number of people entering and exiting an area and determine the number of occupants 
• Generate an alert when an individual is seen in a prohibited area
• When applied to traffic and transportation, provide traffic congestion alerts, illegal parking alerts, and alerts based on different traffic anomalies.

This module can be coupled with the Facial Recognition module to generate an alert if unique individuals enter designated zones.

Video and Thermal Data Streams

This Ready Solution uses the Real Time Streaming Protocol (RTSP) as the streaming protocol for video. The RTSP stream is received and decoded using one or more NVIDIA GPUs, as shown in the following figure: 

Figure 2.              Deep Vision AI modules workflow for video cameras

The configured modules operate on the video stream, generating any corresponding alerts and producing a video stream with overlaid information (such as recognitions and other information on top of each object in the video). You can view these alerts in the Admin Console. 

In addition to supporting a video camera, Deep Vision AI also has a module for thermal analysis, which requires a thermal camera.

The Temperature Detection module identifies the temperatures of people and measures variances to a baseline above or below acceptable thresholds. When people whose temperature falls outside of the set threshold, the module raises alerts. 

A typical camera provides both thermal and video streams. Both these streams are decoded by the NVIDIA T4 GPU. A facial detection module analyzes the video stream (RGB) and the facial coordinates are computed. The temperature measurement module analyzes this information and is mapped to the radiometeric stream, as shown in the following figure: 

Figure 3.              Deep Vision AI module workflow for the thermal camera

Privacy and data protection

Deep Vision AI offers mechanisms to comply with EU General Data Protection Regulation (GDPR) and other international data protection laws: 

• Anonymization of personal data—Deep Vision AI supports a private mode. By using this mode, facial images of nonenrolled subjects are blurred and are not displayed or visible to end users.
• Personal data rights protection—Deep Vision AI offers easy and intuitive mechanisms to remove subjects from watch lists. This option provides users and software operators with full control over personal data. They can act quickly to address personal data removal requests. This protection complies with the rights of subjects to access and rectify data, be forgotten (https://en.wikipedia.org/wiki/Right_to_be_forgotten), and apply restrictions to processing of their personal data and their data portability.