Intelligent Video Analysis System Based on GPU and Distributed Architecture Dr. Shiliang PU Hikivision Research Institute
Challenge in Video Surveillance High Resolution VS storage Complexity VS Accuracy Mass data VS efficiency Mid-size City, about 22,000 cameras 316PB/year Precious video content service under complex situation
Surveillance video content analysis Surveillance video content understanding framework Object Feature Identification detection • Human • Human • Human feature body • Vehicle • Vehicle • Face • others feature • Vehicle
Challenge in Video Surveillance Traditional algorithm can understand simple or standard scene content Traditional algorithm fails in such complex scene content, which is very common in public surveillance. Sun blade closed Glass worn Phone calling Male Safe Belt Teenage Car Height 车型 White Clothes color Ford Fiesta …… 皖 A ?? 66R
Revolution By Deep Learning in Surveillance
Deep Learning in Surveillance Traditional algorithm Deep learning
Deep Learning in Surveillance Pedestrian detection Recall rate, fppi = 0.1 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% overall passenger indoor public sunny rainny winter summer channel area day day Traditional Deep learning
Deep Learning in Surveillance Safe belt Phone Clothes type Riding not fastened calling backpack Hanging bag Mask Hat
Deep Learning in Surveillance Vehicle feature accuracy increased by Deep Learning 100 95 90 85 80 75 70 vehicle color brand model sun blade safe belt phone calling traditional algorithm deep learning
Deep Learning in Surveillance Identity?
Deep Learning in Surveillance Face Recognition Rank in 1 million enroll dataset 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Rank1 Rank10 Rank20 Rank30 Rank40 Rank50 Traditional Deep Learning
Deep Learning in Surveillance Vehicle retrieval based on image comparison
Deep Learning in Surveillance Vehicle Image retrieval 1.0000 0.9000 0.8000 0.7000 0.6000 0.5000 0.4000 0.3000 0.2000 0.1000 0.0000 Rank10 Rank25 Rank50 Rank100 Traditional Deep Learning
These are what we need! BUT……
Limitation on Deep Learning High computing performance Objects detection in surveillance video require 2T Flops/sec, which needs support from high-performance computing hardware. High cost Price of GPU-based server is significant higher than general server. High energy consumption General server costs around 9000KWh per channel every year.
GPU solution based on distributed architecture Tegra Hikvision-Blade Server Base on GPUs
Advantage of Hikvision Blade Server System stability meets industry requirement based on low-cost chip, based on distributed-computing architecture. 1 1-1 1-2 1-3 1-1-1 1-1-2 1-1-3
Advantage of Hikvision Blade Server High performance Low cost Low power consumption Performance/power ratio 18,000 16,000 16,000 14,000 14,000 14,000 12,000 10,000 8050 8,000 6,000 4,000 2,000 550 300 0 Blade Tesla M40*2 General Server Performance ( Gflops ) Power ( W )
Advantage of Hikvision Blade Server Flexibility -for different product forms Smart IPC Sensing Smart NVR Storage Smart Server Application
Intelligent Video compressing standard Smart 264 Background frame General Surveillance Video Video Compressing Compressing Standard Standard IVA Bit rate equalization
Intelligent Video compressing standard 24 hours typical surveillance scene contrast rate at a consistent subjective quality case indoor busy free outdoor busy free H.264 3448Kbps 1715Kbps H.264 1855Kbps 1245Kbps Smart264 945Kbps 315Kbps Smart264 419Kbps 164Kbps Promotion 3.65 5.45 Promotion 4.43 7.57
Intelligent Video compressing standard H.264-3830kbps H.264/H.265 100% H.2651920kbps 50% Smart264-683kbps Smart264 17.8%
Video structured description Human: female male wear glasses riding backpack handbag Vehicle: safe belt fastened/not driver copilot driver’s sun visor c opilot’s sun visor phone calling
Security Big Data framework Other market Cloud analysis 01 Police Traffic High Big data Memory speed Big data service Data mining manager computing data platform bus Fulltext Distributed file database database Cloud storage High speed data bus Non-structured data Structured data Collecting mass data(video, image, alarm, GPS). Extracting structured data from video and images. Offering high speed service, like data searching, analyzing and statistics.
Advantages from Security Big Data 01 Cloud analysis handles mass-data computing problem 02 Big data architecture handles above billions level data Small size Low speed Million Long time 03 Spark memory computing data level cost. Slow offers second degree service feature Low >10 nins extraction accuracy 04 Deep learning increases computing accuracy Issues on traditional system
Security Big Data application City Smart Smart manag …… Police city traffic ement F Q F * 5 8 K O D 1 4 S 6 Y F D A ^ j ! 1 g u f 0 Statistic Analysis Alarm Data inquiry
Security Big Data depth application
Case study – Billion-level image search engine Image search based on image feature extraction and comparison, based on billion-level vehicle images.
Case study- Face recognition system Lost elder found in 5 seconds.
Future Multi-sensor increases data dimensions. Unsupervised learning in video surveillance Optimized neural network framework
THE END HIKVISION Internal
Recommend
More recommend