BioServer Pro 2
  • BioServer Pro 2

BioServer Pro 2

BioServer Pro 2 servers can be deployed flexibly, either on a single machine or in parallel to form a cluster with powerful performance and flexibility. The system is based on the full pipeline deep-learning algorithm and adopts the distributed architecture of heterogeneous parallel computing with GPU as the core.

BioServer Pro 2 facial recognition server mainly provides face detection, feature extraction and face matching retrieval functions, either on photo image under static mode or real-time video stream under dynamic mode. BioServer Pro 2 supports million-scale face images management and high-speed feature extraction. While in static mode, it enables fast large-scale concurrent search with high accuracy of comparison results and controllable number of returned results. Under dynamic mode, the server also featured by real-time facial recognition processing in large-scale real-time video surveillance.

Parameter

Hardware
Dimensions (W*D*H) 448*775.6*87.8mm
Form factor 2U rack server
CPU Intel E5-2600 series *2
GPU GPU card *1
RAM 32GB
Hard drive 2TB enterprise hard drive
Network interface 2* gigabit ethernet port (RJ45)
Power 1+1 redundant power system
Weight 35kg
System Environment
Operating system CentOS 7.2
Runtime environment CUDA 8.0, Redis, MongoDB
System Parameters under Dynamic Mode
Total target capacity 1,000,000 templates
Single target library capacity Maximum 200,000 templates
Number of target libraries Maximum 200
Capacity of target library for real-time surveillance Maximum 200,000 templates
Video decoding GPU decoding
Video processin capability 8 channels
Video input standard
The compression standard of input videos shall be H.264.
1. Support GB/T28181 and RTSP video stream
2. Support offline video files
3. Support third-party video platforms
Cache capacity for captured facial picture
Size of each picture shall not exceed 200kB.
When 85% of storage space of the hard drive is used, old pictures will automatically be overwritten.
Batch-import speed of templates 1,250,000 pictures/hour
Import speed of single template Real-time monitoring: 20 ms/picture;snapshots: 4 ms/picture
Result’s return time for real-time facial capturing <1 second
Result’s return time for real-time facial comparison <1 second
Number of results returned after real-time facial comparison <10
System Parameters under Static Mode
Total target capacity 1,000,000 templates
Single target library capacity Maximum 200,000 templates
Number of target libraries Maximum 200
Batch-import speed of templates 1,250,000 pictures/hour
Import speed of single template 4 ms/template
Speed of retrieving features in template batches 350 templates/second
Speed of retrieving features in single template 3 ms/template
1:N image concurrent search volume
90 pictures/second
(return top 100 results)
1:N feature concurrent search volume
200 pictures/second
(return top 100 results)
1:1 image concurrent comparison 125 pictures/second
1:1 feature concurrent comparison 600 pictures/second
Algorithm Parameters
Maximum number of detected faces per frame 30
Maximum number of recognized faces per frame 20
Pose angle tolerance for face detection ±75° (yaw); ±60° (pitch);±45° (roll)
Pose angle tolerance for facial recognition ±30° (yaw); ±30° (pitch);±30° (roll)
Minimum face size for capturing 20 * 20 pixel
Minimum face size for recognition >30 pixel between eyes (optimized: >50 pixel)
Dynamic facial attributions Age, gender, face shapes, glasses,sunglasses, smile, masks, races,stares, mouth opening, beards,hats
Facial recognition accuracy rate Dynamic mode:
94% accuracy rate with 10% error rate
85% accuracy rate with 1% error rate
Static mode:
90% hit rate for Top 1 result, 98.5% hit
rate within Top 5 results (under 1
million-scale 1:N comparison)
Environmental Requirements
Temperature requirements
Working: 10 ~ 30 °C (50 ~ 86 °F)
Storage: -40 ~ 55°C (-40 ~ 131°F)
Humidity requirements
Working: 35 ~ 80%
Storage: 20 ~ 93 %