Research On Negative Emotion Recogniton Of Automobile Drivers Based On Image Processing

The operation of drivers in negative mood is easy to produce dangerous driving,which leads to traffic safety hazards,such as forced congestion,swearing and provocation of drivers in anger,rapid heartbeat,dull eyes and slow reaction in fear,etc.Most of the existing recognition methods for driver’s negative emotions use facial expression changes or physiological parameters changes,but there is no combination of the two.In view of this,this thesis proposes a non-contact recognition algorithm of driver’s negative mood,which combines the change of driver’s facial expression with the change of heart rate.The algorithm obtains the driver’s facial image through the camera,and recognizes the change of driver’s facial expression and heart rate through image processing.The two are combined by different weights to realize the driver’s non-contact recognition of negative mood.1.In order to identify the driver’s negative mood,this thesis sets different weights for the changes of facial expression and heart rate,and compares the output results with the set threshold to find out whether the driver is in negative mood.2.For the recognition of driver’s facial expression,this thesis proposes an expression recognition algorithm based on local candidate regions.The algorithm is based on 68 feature points of face and divides the face into 8 local candidate regions according to the facial muscle movement.Then the features are extracted in each region by parallel network.According to the area ratio of the expression and the local candidate region,the expression is divided into two kinds of expressions: positive and negative.For the driver’s body heart rate parameters,when there are negative emotions,the heart rate data will suddenly change.3.The heart rate detection method is based on Romote Photo Plenthysmo Graph(RPPG)technology,through the camera,based on 68 feature points of human face,select part of the forehead and nose as two candidate areas.Then divide them into several small blocks,through the k-means clustering green channel mean,through the subsequent Butterworth band-pass filter,Fast Fourier Transform(FFT),get rough unstable heart rate value,through the Kalman filter,Long Short-Term Memory(LSTM)get stable heart rate.Then the change of heart rate is identified by neural network.The experimental results show that the recognition precision of the algorithm reaches93.40% in the data set,which is 4.00% higher than that of the algorithm composed of single facial expression recognition.The facial expression recognition algorithm proposed in this thesis achieves 99.85%,96.61% and 99.06% precision rates on the open expression data sets CK +,JAFFE and self-defined FED data sets,which is 6.01%,10.17% and 6.76%higher than the same period expression recognition S-Patches algorithm.For the recognition of heart rate increase,the algorithm achieves 100% precision on the test set.The algorithm provides an important idea for the study of driver’s negative mood recognition,and provides a certain guarantee for driving safety.