Research Of Driver Emotion Recognition Under Partial Occlusion

With the advent of the Industrial 4.0 and the development of science and technology,intelligent driving system is booming,and people pay more and more attention to driving safety.For drivers,most traffic accidents are caused by emotion fluctuations,so it is very important to accurately identify the driver’s emotion.For the task of emotion recognition,good accuracy has been achieved under controllable experimental conditions which mean the complete frontal face.This means that emotion recognition is not satisfactory in the uncontrolled natural environment with different illumination,posture,resolution and occlusion.At the same time,emotion recognition under partial occlusion has not received much attention.Drivers will inevitably wear sunglasses,masks,face towels or some other obstructions.It is very important to study the driver’s emotion recognition under these obstructions.In this paper,the driver’s emotion recognition under partial occlusion is studied.Some exploration,research and expansion of face detection and emotion recognition algorithms are carried out aiming at detecting the driver’s face under occlusion and improving the accuracy of emotion recognition,and detailed experiments and analysis are carried out.The specific research is as follows:（1）Morphological operations are carried out on the driver face image to solve noise problem,and the image is preprocessed for face detection and emotion recognition.（2）By studying the driver’s face detection under occlusion,especially with glasses or towels,a face detection method under partial occlusion based on Gabor wavelet transform combined with occlusion mask and improved LBP feature is proposed.（3）A network of facial emotion recognition based on attention mechanism is proposed which improves the pooling operation of attention module.At the same time,multiple cross attention heads are used to make the network focus on multiple facial regions rather than the whole driver’s facial information,so as to reduce the impact of occlusion area on emotion recognition.The experimental results show that the face detection algorithm proposed in this paper can effectively detect the driver’s face under partial occlusion.The detection rates of non-occlusion,eye-occlusion and mouth-occlusion in AR data sets are 99.2%,95.79%and 87.05% respectively.The emotion recognition algorithm proposed in this paper can also effectively deal with the emotion recognition under uncontrolled conditions,and the recognition accuracy is 89.1% in RAF-DB data set.