Abstract
Gaze estimation is one of the current important research contents of computer vision. For the current situation where the gaze estimation neural network has a large amount of parameters but the accuracy is not greatly improved and the head pose is difficult to handle, this paper proposes a simplified gaze estimation network model SLeNet based on the LeNet neural network. The deep separable convolution in the Xception network is used to reduce the amount of parameters in the convolution part and improve the computational performance of the network model. The method of splicing head posture features is retained, but another branch neural network is designed to learn head posture based on eye image and mouth corner information, and no additional module is required to obtain head posture separately. The improved network model is used to compare experiments with the original network and VGG-16 on the MPIIGaze dataset. The results show that the improved SLeNet network model performs better on the MPIIGaze dataset than LeNet and VGG-16 and has fewer parameters.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call