Abstract:
Smart digital advertising requires face detection as the initial stage to recognize the person’s attributes by localizing human facial areas. This technology tends to ope...Show MoreMetadata
Abstract:
Smart digital advertising requires face detection as the initial stage to recognize the person’s attributes by localizing human facial areas. This technology tends to operate with CPU-based systems. The Deep Convolutional Neural Network approach has demonstrated excellent accuracy for face detection work. However, this architecture involves heavy computations and parameters because it uses many filter operations. It causes deep architecture to slow down the detector speed. Moreover, a practical application entails using a detector that can operate in real-time. The recent CPU-based face detectors operate slowly in an integrated system. This study proposes a faster face detector to predict the human face area using efficient architecture robustly. The architecture consists of a light backbone to discriminate distinctive features and a four detection module to predict multiple faces. In order to bridge the three prediction layers, it implements a high-level transition module with a cheap operation. It also offers a new light attentive block to highlight typical facial features at each detection module efficiently. As a result, this detector achieves excellent performance and outperforms other low computing detectors. The proposed detector can fast operate at 112 frames per second on a Core I5 CPU and at 11 frames per second on a Lattepanda device, faster than other competitors.
Date of Conference: 11-15 July 2022
Date Added to IEEE Xplore: 25 August 2022
ISBN Information: