Analysis And Research On Operator's Working State Based On Video

With the development of imaging device,computer vision technique has been widely used in various industries.Along with the development of high technology and the progress of industry,there are also stricter requirements on operator status.If the operator is not concentrated,or have the symptoms of fatigue and sleepiness will affect the operation of the instrument or vehicle,endangering property and even life safety.Therefore,it is of great theoretical value and practical significance to use the camera to monitor the operator and extract the corresponding features from the operator's facial expression to judge the operator's status.This paper first introduces the relevant theories of the operator's working state detection.Through analysis and research,the state of the operator is judged by the method of computer vision,and the face detection technology and facial landmark detection technology are used as the basis for the operator state detection judgment.By reading the literature and analyzing the existing face detection algorithm and facial landmark detection algorithm and the judgment indicators of the operator state detection,the face detection algorithm,facial landmark detection algorithm and state evaluation indicators suitable for the actual application requirements of this article are selected.In this paper,according to the real-time and reliability and requirements for equipment resources,MTCNN network is selected as the method of face detection and facial landmark detection.A method for extracting the eye region was designed based on the marking of the key regions of the face with the facial landmark detection,and the mouth region was extracted as the basis for further judging the state of the eye region and the mouth region.The trajectory of head movement in real space is obtained by using the coordinate transformation of the key points of the key parts of the head,so as to judge whether there is abnormal head movement,and then judge the status of the operator.Based on the facial landmark detection technology,a new eye and mouth state detection method is proposed,which combines the aspect ratio of the eye and mouth areas and the image features of different states to improve the eye and mouth state detection.While ensuring real-time performance,the detection accuracy is improved.And in the eye area,the pupil position is detected.Finally,multiple features such as eyes,mouth,pupils,and head are fused to detect the state of the operator.Experiments show that the method proposed in this paper has good accuracy and efficiency in the state judgment of operators,and is more stable than the state judgment of a single feature.