Power Learning Based On Machine Learning Research On Personnel Behavior Recognition

With the progress of science and technology and economic development,the national to the requirement of increasing the quality of life,the power requirement also gradually improve,so the state grid to the scene of the electric power operation and administration of production safety but also pay more attention to related department supervision,management management of dereliction of duty and power operation personnel itself to don't follow the rules and regulations,easily lead to big safety accident happened because of the traditional safety supervision mode for human Therefore,in order to ensure the orderly supply of the power system and the stable operation of the power system,the state grid and all power units must change the inherent safety supervision mode and adopt the automatic and controllable supervision mode.This paper focuses on the two tasks of the local discharge and distribution variation experiment in power operation site: equipment grounding detection task and forklift truck personnel assisting detection task.The main research contents and contributions are as follows:1.Two methods of human key points extraction are introduced: top-down human key points detection and bottom-up human key points detection,and the bottom-up human key points detection algorithm based on openpose is introduced in detail.The theory of support vector machine(SVM)is introduced in detail,which lays a foundation for behavior classification.This paper briefly introduces the common tracking algorithm,which is the foreshadow for forklift truck tracking.2.The purpose and method flow of the grounding detection task are described in detail.A three-frame difference method based on HSV color feature was proposed to detect the movement of the grounding rod,and openpose algorithm was used to extract the key points of the personnel,and a two-dimensional human model was established.By analyzing the behavior features,feature vectors were extracted,and SVM was used to classify the behavior related to the equipment grounding detection task.Finally,the feasibility of the algorithm flow is verified by experiments.3.The purpose and method flow of the forklift truck personnel assisted detection task are described in detail.A multi-feature fusion CamShift tracking algorithm wasproposed to reduce the influence of illumination by establishing h-s 2d histogram.By adding Harris corner feature,the visual deformation error caused by forklift truck during turning is reduced.The Kalman prediction mechanism was introduced to improve the recognition rate of the forklift truck when the forklift truck was blocked.This paper proposes a human assistance judgment algorithm,which USES the multi-feature fusion CamShift tracking algorithm to extract the forklift location information and the human location information extracted by YOLOv3.Finally,the feasibility of the algorithm flow is verified by experiments.