Recent Advances in GPU-Based Intelligent Video Analysis Beijing - PowerPoint PPT Presentation

GTC'16, San Jose, USA April 5, 2016 Recent Advances in GPU-Based Intelligent Video Analysis Beijing Vion Technology Inc. Hai Tao, Dr., CEO

Company Introdution • The Beijing Vion Technology, Inc.(BVT) was founded in 2005, specialized in computer vision technology. The company currently employs around 200 talented staff members. San Jose, USA April 5, 2016 • The company has developed products in several markets including intelligent transportation systems (ITS TS),business intelligence systems, security and urban management systems. These products are deployed broadly across China and BVT is recognized as one of the leading computer vision companies in China. • Recently, BVT has developed core algorithms for ADAS and UBI products.

A Typical IVA System PC Client San Jose, USA April 5, 2016 Analog Cam Switch Data Manage IP Cam ment Network 数据中心 IVA Box Half Dome Cam Concealed Camera Mobile App

TK1-Based IVA Box - VT-B2081A (CBOX) San Jose, USA April 5, 2016

TK1-Based IVA Box - VT-B208 2A (IBOX)  2 TK1 GPU's San Jose, USA  8 analog video inputs April 5, 2016  8 audio inputs  2 RJ45 1000M ethernet ports  built-in GPS  1 2.5" hard drive  Alarm: 4 in, 2 out  1 RS232 and 1 RS485  SPF connection

DEMO: IVA Algorithms Running on CBOX/IBox • People counting - Shopping mall • People counting - Summer Palace San Jose, USA April 5, 2016 • People counting - Bus • People counting - Subway • People counting - Cinema • Urban management - Vending • Law enforcement - Fighting detection • Law enforcement - Parking violation

A GPU-Based Smart Camera - VT-E412 San Jose, USA April 5, 2016

A 12MP@30fps GPU smart camera  The 1st GPU-based smart camera that can reach 12MP @30 fps  Applications: Face recognition, gender and age recognition, people counting, LPR, fight and chasing detection, premiter monitoring  Main features  12 MP CMOS sensor @ 30 fps  FPGA+TK1 ISP pipeline  Full resolution video analytics  Sensors: gyroscope, accelerometer, digital compass  GPS & Wifi/3G/4G supported  Audio: 1 in, 1 out  Storage: 2 SD cards  Alarm: 2 in, 1 out, 1 RS232 and 1 RS485

DEMO: IVA Algorithms Running on VT-E412 • Facial detection • Gender and age recognition San Jose, USA April 5, 2016 • Vehicle detection, tracking and recognition

San Jose, USA April 5, 2016 StartNet GPU Cluster

40 GPU s in a 2U Rack Server  Characteristics of backend IVA  Handreds of h.264/HEVC video streams from IPCs San Jose, USA April 5, 2016  The computation need for decoding and vision algorithm roughly in 1:1 to 1:10 range  Using traditional server: not enough matching HW decoding power, higher FLOPS/Watt, Lower computing density (FLOPS/SpaceUnit)  Many-GPU solution  TK1/TX1 balances the HW video decoding/encoding capabilities with computing power  Lower FLOPS/Watt  Much higher computing density

Vehicle barnd/model and color recognition  Frontal and back view of vehicles San Jose, USA  Day/night time April 5, 2016  Average 90% accuracy for vehicle brand recognition

Driver behavior recognition  Not fastening seatbelt San Jose, USA  Using cell phones April 5, 2016

Prototype ADAS Algorithm Framework  Deep neural network based object detection and tracking (vehicles, pedestrians, bicycles, etc) San Jose, USA  Deep neural network based road scene understanding (traffic April 5, 2016 lane detection and following)  License plate recognition  ADAS main functions including lane departure warning (LDW),vehicle distance monitoing warning (HMW), pedestrian collision warning (PCW), and driving behavior analysis (DBA)

Prototype ADAS Algorithm Framework  Demonstration San Jose, USA April 5, 2016

San Jose, USA April 5, 2016 Thank you!