Deep convolutional neural network-based Bernoulli heatmap for head pose estimation

Head pose estimation is a crucial problem for many tasks, such as driver attention, fatigue detection, and human behaviour analysis. It well known that neural networks are better at handling classification problems than regression problems. an extremely nonlinear process to let the network output angle value directly optimization learning, weight constraint of loss function will be relatively weak. This paper proposes novel Bernoulli heatmap head from single RGB image. Our method can achieve positioning area while estimating angles head. The makes it possible construct fully convolutional without connected layers provides new idea form estimation. A deep (CNN) structure with multiscale representations adopted maintain high-resolution information low-resolution in parallel. kind rich, representations. In addition, channelwise fusion make weights learnable instead simple addition equal weights. As result, spatially more precise potentially accurate. effectiveness proposed empirically demonstrated by comparing other state-of-the-art methods on public datasets.