Low cost embedded solution for facial capture, tracking, and recognition all-in-one machine

l Adapt to multiple lighting conditions and scenes from various angles

l Automatically select the best facial photos for comparison

l Facial detection rate ≥98%;Adapt to up, down, left, and right deviation ±60Degree; Facial recognition accuracy ≥97%

l Comparison of witnesses1：1Passing rate ≥99%;Capture real faces1：N（N=5000(Regular version)or200000(High end version)）Passing rate ≥97%Million levelFirst to middle rate of ID card database ≥90%;On site photo recognition rate ≥99%

l 200Ten thousand pixel image, average processing time ≤100ms

l 1sImplement facial comparison internally

l Effective recognition distance support3-20mSimultaneously recognizing the number of people in the region is supported7-10people

l All in one machine supports 5000Personal face whitelist or blacklist (regular version), 20000Personal face whitelist or blacklist (high-end version)

l Support offline headcount statistics, gender recognition, age recognition

l Support offline black and white list alarm

l Support NetworkNetwork output485output

Core Features of Zhonghui Face Comparison Camera

l The face detection technology based on improved multi task cascaded convolutional neural network reduces the requirement for image quality and significantly improves the speed of face detection

l Based on improved optical flow, face tracking technology is used to quickly track faces in video sequences. Through face evaluation methods, only high-quality face images are selected for face recognition, effectively reducing the system's load

l Based on improved residual network(ResNet)The facial recognition technology utilizes improved alignment processing on high-quality facial imagesResNetExtracting facial features from the network and calculating their similarity to faces in the database

l The latest processor is equipped with high performanceGPU+CPUParallel hybrid distributed architecture for improved performance5-10Double, bringing super computing power to complex mathematical and geometric calculations

l Deep learning models are trained from billions of samples and have excellent generalization performance

l All processing is completed in an all-in-one machine, without the need for additional desktop computers or servers

l Strong scalability, easy to develop multiple application functions based on recognition results

l The recognition results of the all-in-one machine can be centralized through the network [optional], providing support for big data analysis