Sightcorp deep learning method improves biometric face detection at challenging angles

Artificial intelligence software company Sightcorp has developed improved biometric facial detection capabilities for facial analysis and recognition software based on deep learning, rather than traditional Haar Cascade detector methodology, the company has announced.

Sightcorp creatively iterated its face detection technology after an effort to enhance its software to provide deeper insight into moment-to-moment interaction. The company’s software is intended to provide users with as much insight as they need to make informed, predictive decisions, with data such as subject’s emotions, demographics, and attention spans.

In order to make AI-powered face detection effective for a wide variety of applications, Sightcorp focussed on deepening the ability of the software to detect faces in diverse head poses with greater accuracy, speed, and granularity. Between different resolutions, light conditions, camera angles, and even facial coverings, existing systems often have difficulty detecting faces, according to the announcement.

The annotations and indications of head pose in most modern datasets provided a baseline for Sightcorp to work from, and the company used the WIDER FACE and VGGFace datasets to determine that it would need to quantify detection performance across granular variations in yaw, pitch, and roll.

The team then used the Head Pose Image Database from the Prima Project at INRIA Rhone-Alpes, which contains 2790 monocular face images of 15 people from a range of pan and tilt angles. The team engineered a small Python script for a ground truth CSV file with a path to each file in the dataset, along with its pitch and yaw values, and created a benchmark for the AI algorithm to take over and run its own learning process.

Sightcorp also created a heatmap to compare the face detection of its method against a traditional Haar Cascade method, and found that its new method provides a significant advantage in detecting faces captured from the side, above or below.

“I am very proud of what the team has achieved with this face detection model,” comments Sightcorp CEO Joyce Caradonna. “This improvement doesn’t only enhance the software’s core promise, it makes a shift in the way datasets are being used, giving the next generation of comparable analytics platforms that are sure to emerge, a new baseline from which to innovate.”

Methods for improving face detection and recognition at different camera angles have also been developed by other companies including ISS and Neurotechnology.

Article Topics

artificial intelligence  |  biometrics  |  deep learning  |  face detection  |  Sightcorp