WiMi Hologram Cloud Inc., a leading global Hologram Augmented Reality Technology provider, today announced the development of an AI chip system for holographic face recognition based on edge computing. The system places recognition, acquisition, and analysis at the terminal, effectively improving the optimization of algorithms for one hand and allowing the establishment of a private domain to protect data security for another effectively. This system can be used in some key departments and enterprises or in controlling high-end factories and the security management of industrial parks, office buildings, flats, etc. It is easy and safe to deploy with high efficiency.
The system is different from traditional face recognition and ID match. It enables simultaneous face-tracking acquisition and face attribute analysis feedback results. It acquires holographic high-density face data of the subject at the front end: attributes, appearance, features, collection time, geographical location, and other essential information. These features can be identified and distinguished. WiMi’s system combines edge computing, AI arithmetic acceleration, deep learning algorithms, holographic data gain technology, convolutional neural networks, face recognition, and acquisition to merge and upgrade the existing security video system.
WiMi’s system, which uses a time window for the sampling period, can also be set up according to specific targets, such as critical positions, time on duty, and geographic space. If multiple samples are in a sampling period, the system will select the best sample as the last sampled information. The system will recognize all faces if multiple faces appear in a sampling frame. For completely unrecognizable looks, the system adopts an ignore-and-remain strategy. It focuses on capturing them again in subsequent frames or other surveillance cameras until they are recognized, thus ensuring full recognition and data integrity and keeping the area safe.
This system can provide structured primary data for security management and production safety by matching high-density dynamic personnel information collection to surveillance video of key locations.
The system includes a video access port, a hologram decoding and frame extraction module, a hologram optimization module, an edge computing and algorithm acceleration module, a face acquisition and analysis module, central control module, and a data storage and notification module.
Video access port interfaces with existing surveillance video. Hologram decoding and frame extraction module deals with the frames according to the central control module. The hologram optimization module performs image acquisition and analysis in the extracted frames, image optimization acceleration of the sampled frames, and sends feedback to the central control module. If information gets lost, new instructions will be issued by the central control module to reproduce information. With ARM architecture, the edge computing and algorithm acceleration module contains core computing units. Embedded with multi-layer CNNs, the computing unit performs algorithmic operations on low and high parallel computing performance. The face acquisition and analysis module collects recognition, segmentation, and extraction data by combining photos, geographic information, and time information. The collected face information is analyzed for attributes, gender, age, ethnicity, masks, and glasses. The central control module realizes the sampling process for management, integrated control, and management of other modules. The data storage and notification module stores the collected personal information locally and can notify external systems according to the information level.
The system’s front end accesses the video through a dynamic holographic face recognition algorithm based on edge computing, decodes the video holographically, and detects, tracks, captures, and de-emphasizes the faces in the picture. The system uses feature values as information identifiers to build information on pedestrians, completing the collection of information on people and enabling private domain management to improve information security levels. In addition, the system is easy to deploy. It can be deployed in various ways, including external, rack-mounted, and mobile. The device can be directly connected to existing HD network cameras. It can be used directly on the front end to complete part of the video structuring work, obtain high-quality face-structured data, and improve the speed and calculation efficiency of back-end intelligent identification and analysis, making full use of the existing stock of cameras. Existing unstructured video can be directly upgraded to smart structured data through external attachments.
WiMi’s edge computing-based holographic face recognition AI chip system can be used in various essential situations as it can perform high-density dynamic holographic face capture in complex environments. The system uses a digital camera with an intelligent front-end for face information collection and can meet a wide range of requirements for security information collection.