PENSEES

PES-Brain

Cloud Intelligence
PES-Brain
Edge Intelligence

Original Technologies

Image/target Processing

Video Structuring

Video Summarization Description

Automatic Driving

Intelligent Robot

Speech Recognition

AI Intelligent Sensor

Enhanced Image Restoration

Image/Target

Detection-tracking-recognition-classification
Moving object detection and tracking: fast detection and tracking of moving objects in video images.
Object detection-classification-recognition-segmentation
Accurately and quickly detect the target in the scene. Classify and identify, and then carry out semantic and instance segmentation under a variety of complex environments.
Face detection - tracking recognition anti-counterfeit
Fast and accurate face detection, tracking, quality judgment, key detection alignment, face feature extraction, face recognition of different ages, different scenes, and RGB/IR face anti-counterfeiting under unconstrained conditions.
Pedestrian detection, tracking, recognition
Fast and accurate detection of pedestrians, tracking, quality assessment, skeleton key points extraction, pedestrian feature extraction, pedestrian recognition/ReID under unconstrained conditions.
Vehicle/non-vehicle detection tracking recognition
Fast and accurate detection of vehicle, motorcycle, tricycle, bicycle, tracking, quality assessment, key points extraction, vehicle feature extraction, vehicle recognition Car ReID under unconstrained conditions.
Human vehicle behavior analysis - anomaly detection and recognition
Fast and accurate analysis of the behavior of people/cars under unconstrained conditions, tracking, and monitoring of abnormal behavior, including fights, falls, clustering, theft etc. and different detection, analysis, identification required by different scenes.

Video Structuring

Human, vehicle and object detection and tracking
Achieve accurate and fast detection and tracking of human faces, human bodies, vehicles and objects under unconstrained conditions.
Description of faces and pedestrians features
By analyzing the physical and structural characteristics of faces and pedestrians, we are able to accurately identify and decode up to 27 different categories of face and human body features in videos and images, and another 128 sub-categories.
Description of vehicle and non motor vehicle features
By analyzing the structural characteristics of vehicles and non-motorized vehicles, we are able to identify and decode up to 22 different motor vehicles, 27 types of license plates and 5500 types of vehicle models; another 3 categories of non-motor vehicle features and 18 sub-categories.

Video Summarization and Description

Dynamic background extraction
By using the latest image processing and deep learning technologies, the separation, extraction and fusion of background, static objects, moving objects in different scenes can be realized.
Moving/abnormal target detection and tracking
Moving/abnormal target detection, tracking, trajectory and salient content extraction.
Video intelligent compression
Intelligent extraction and storage of significant and important video content from ultra-long videos.

Automatic Driving

Visual perception cognition
Machine vision, robot vision input, including sensors and light source design, information processing and analysis, image processing, machine learning and deep learning, and making judgments with various empirical knowledge.
Path planning - intelligent control - reinforcement learning
Based on accurate perception results, it realizes scientific and humanized decision logic, plans reasonable driving paths, and provides precise vehicle control function.
Intelligent navigation-Visual SLAM
SLAM builds a map via positioning and realizes intelligent navigation to achieve path planning and motion control for mobile robot, mechanical back movement and other motion planning.

Intelligent Robot

3D reconstruction
Design scenes of different cameras shooting objects or surroundings, the objects or surroundings will be presented in three-dimensional vision through optical geometry, computer vision, machine learning, deep learning, online or offline rendering.
Human-computer interaction
Design computer-based robots to identify and understand certain human language actions, and respond in natural language and behavior.
Augmented Reality
Integrate technologies including SLAM, 3D modeling, real-time tracking of target detection, scenes fusion, computer graphics, and other technologies to enhance reality.

Speech Recognition

Voiceprint recognition
The world's cutting-edge technology of "hearing and recognizing people" , through the analysis of the characteristics of one or more speech signals, the unknown voice can be identified and the identity of the speaker can never be discriminate.

AI Intelligent Sensor

Computational optical imaging technology
It is an emerging interdisciplinary technology involving geometric optics, information optics, computational optics and modern signal processing, which can be accurately described by mathematical models.
AI image quality assessment
Use AI deep learning, reinforcement learning to automatically give feedback to pictures or target quality so that the best picture or video is taken under any condition.
AI-ISP capture
AI algorithm is used to effectively replace the traditional ISP (Image Signal Processing).
Sensor feature fusion
By adopting multiple sensors features fusion at the cameras side, the computer visual application accuracy and special needs are increased.
Human face detection and anti-tampering detection
Identify real faces from the cameras end, and alter the tampered scene.

Enhanced Image Restoration

Super resolution
By using the latest technology of image processing and deep learning, we can improve the resolution of pictures without losing the quality of photos and present clearer content.
Reduce noise, blurring, jitter
Based on leading image processing algorithm, fast reduction of camera noise, blurring, jitter can be realized.
Fog-rain-snow removal
Based on the industry's leading image processing algorithm, the photos that are disturbed by fog, rain and snow can be adjusted with higher resolution.