| Human pose estimation is a method to obtain specific attitude by detecting the human body in the image or video information.It is the basis of many human-computer interaction tasks.In the field of intelligent safe driving,driver attitude estimation technology has a wide application prospect.In this paper,two human pose estimation networks are designed based on convolutional neural network for the complex environment in the driving room.The main contents are as follows:The FCNfs network model be proposed in view of the influence of common lighting changes in the driving room and similar parts of human body in the background on the accuracy of driver attitude estimation.The network can simultaneously detect the keypoints and the partial Affinity Fields between the human body’s keypoints,and finally use the Part Affinity Fields(PAFs)algorithm to match and connect the detected keypoints.Dataset for the driver’s driving situation(DDS)was constructed using the evaluation criteria consistent with the COCO dataset.Finally,compared with other advanced algorithms on the COCO dataset and DDS dataset,the experimental results show that the model can effectively improve the detection accuracy of the joint.The transformated-hourglass RMPE Network model be proposed to avoid the influence of non-target regions on the detection results of the keypoints.The top-down method is used to detect the human body,and then to estimate the posture of the human body in the detection area.The model uses the 2-stage transformated-hourglass Network to realize the process of feature detection from rough to fine,and adopts the strategy of multi-scale feature fusion across stages to make the feature graph contain more spatial information.Finally,the experimental results show that the model can effectively improve the performance of attitude estimation.The paper has 29 pictures,7 tables,and 55 references. |
