People recognition

People recognition software

People recognition is the most demanded technology in computer vision

Humans pay high attention to information about themselves and their fellow beings, so there’s no surprise that people detection, tracking, counting, recognition, and identification are the most demanded technologies in the whole computer vision domain.

Historically video surveillance and biometrics were the drivers for technologies that allow extracting a person’s image from the whole footage and identifying him. Before this information was used solely for security purposes. However, today companies, as well as numerous state agencies, have started using information about people as a part of business intelligence. For example, city agencies collect information about pedestrian traffic flows; retailers count their visitors; internet companies detect people and faces in uploaded images to improve search results; marketers identify returning visitors; advertisers measure people’s emotions, etc.

Today people recognition software detects body poses, faces, appearance features

Detecting people in the video stream started gaining popularity at the beginning of the 2000s as computers came to surveillance. These first detectors worked as the simplest object detectors that tried to measure pixel dimensions of moving an object on a static background. Object with size within certain boundaries was considered as a person, regardless of actual shapes that fit in these boundaries. Recent progress in computer vision and machine learning lets people detectors significantly increase their performance. A modern well-trained neural network can detect a human body in any pose, at minimal image quality, on any background. Moreover, it can also extract appearance features that allow visually distinguish and track persons in the crowd.

In part of facial detectors, deep convolutional neural networks also brought recognition reliability to an absolutely new level. Previously facial recognition relied on hard-coded extraction of facial features, used as a facial representation. By continuous efforts of more than a decade, methods improved the recognition accuracy of the famous LFW benchmark to about 80%. This figure was valid only for very constrained face capturing and still couldn’t be applied to the most real-world applications due to its unstable performance.

Today facial features are not hard-coded, but “learned” by a deep neural network. State-of-the-art accuracy of face recognition systems exceeds 99.7% now, which is better than trained human performance (97.5%). In addition to person identification, face recognition modules can relatively accurate estimate people’s gender, age, gaze direction, expressed emotion. Various types of companies already found efficient and sometimes even elegant ways how to use automatically captured people’s features to improve their business performance.

Automated people recognition converts CCTV into business intelligence

BitRefine offers many person-related neural recognition modules for its multi-purpose object recognition platform BitRefine Heads: people detection, pose recognition, face recognition, gender recognition, age recognition, emotion recognition, facial features extractor, etc. Each module extracts corresponding data from given video or images and saves them to the DB for further use. Users can choose any of the available modules, use it independently, or add several of them at once to the processing pipeline to get the object’s various properties.

In addition to people detection and recognition modules, BitRefine offers advanced tracking modules that can effectively deal with crowds and tight people flows. Modular architecture allows adding further standard object-related modules, such as object’s color detector, counters, etc. BitRefine Heads platform is designed with the main purpose of converting the client’s raw video footage into meaningful information, that the client can (and should) leverage on. Information about people and their properties is one of the most obvious examples of such insight.

Related Solutions

Pedestrian traffic counter

Tracking people effectively depends on the settings. For instance, tracking people on busy streets round-the-clock is not the same as tracking ...

Retail video analytics

As a data science company, the BitRefine group helps retailers solve such tasks as precise pricing, targeted advertising, demand forecasting, ...

Face identification system

Face recognition has been the prominent biometric technique for identity authentication. In the 1990s researches achieved significant success in the ...

We’ve got more video analytics to share. Let’s be friends. But you first...