AI Intelligent Traffic Big Data Structured Video Anti - Terrorism Security

Product Specification

AI, intelligent traffic, traffic, big data, structured video

Product Overview

With the booming trend of video surveillance cameras in last decade, high-definition networking video surveillance systems have been widely used for auxiliary management and anti-terrorism security over the world. In case of incidence, however, the monitoring process of the surveillance video is time-consuming and labor-intensive, whether it is event precaution, real-time monitoring in the matter, or security analysis in hindsight. Furthermore, large amount of video data are lost without effective information extraction, resulting in the inefficiency of the existing video surveillance systems.

Due to the urgent need of an automatic video analysis system, We have developed “GTVision”, which is a video content structuring and analysis system as intelligent as the human vision system. As the core system of government supervision, enterprise management and social security, it can extract the useful information during the whole surveillance process without losing any clue for event analysis. This system is an important information module for the infrastructure of smart city, smart transportation and smart security.

In the field of smart transportation, GTVision has advantage to manual monitoring as it is able to recognize and count motor vehicles, non-motor vehicles and pedestrians on all existing roads at an overall accuracy of above 90%, on a 24 hours by 7 days basis. By extracting the information from the video content of all roads and blocks, a group of census data can be acquired, such as traffic directions of all vehicle types, motion trajectory, and congestion level. All these data are beneficial to effective decision of a smart transportation system.

Main Function

Fundamental research of the cutting-edge technologies in machine learning. Top level in the world.

Face recognition, Crowd density monitoring, Cross camera tracking, Vehicle flow monitoring, Traffic accident monitoring, Urban traffic scheduling.

Pedestrian, Vehicle, Ever increasing amount of data, It can display real-time traffic visualization and machine learning efficiency on LCD wall

Through the big data analysis, the system can provide traffic light management department with time plan suggestions and vehicle priority recommendations for traffic lights.

By setting up special monitoring camera and other equipment at traffic intersections, traffic data can be captured, and real-time traffic flow and speed data can be recorded through image analysis technology.

Traffic big data, can provide statistical analysis results for MMDA scientific adjustment of traffic lights, optimize road management, and provide scientific basis for the construction of viaducts and tunnels.

Product Specification

Hardware Requirements:

CPU: Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz

Memory: 64G

SSD: 512G

HDD: 4T

GPU: GeForce GTX 1080Ti 11G

Object Detection:

Minimum required resolution: 1280x720 (720p)

Precision rate: >90%

False positive rate: <10%

Maximum detected objects in a frame: 600

Minimum resolution of detected objects: 8*8 pixels

Deep learning network layers: ~200 layers

TPS by single GPU:

1080Ti: 75 fps, or 3 cameras (720P) full frame detection

P40: 100 fps, or 4 cameras (720P) full frame detection

P100: 125 fps, or 5 cameras (720P) full frame detection

Object Tracking across Cameras:

Accuracy under single camera: 90%

Accuracy across cameras : 80%

Simultaneous targets tracking number limit under single camera: 100

Minimum resolution of target: 8*8 pixels

TPS (throughput per second) by 4 GPUs: 300 frames (720p) per second or 12 cameras (720P) full frame detection

Image Retrieval:

Top-1 accuracy rate: 80%

Top-10 accuracy rate: 90%

Single query response time upon million-level database: <100 milliseconds

Single query response time upon billion-level: <20 seconds