Detection And Tracking Of Safety Helmet In Scene Of Thermal Power Plant

In the dangerous working environment,the safety helmet plays a vital role in protecting the head of the staff.The traditional manual monitoring method needs a lot of manpower,and the monitoring personnel are easy to fatigue,easy to appear false detection.The detection and tracking of safety helmet by video automatic monitoring can reduce manpower and accurately monitor the safety helmet wearing of workers in thermal power plants,so as to reduce the probability of accident hazard.When the safety helmet is tested,the safety helmet worn by the workers in the perspective is not easy to be tested due to the complex environment and large area in the thermal power plant,which leads to the problem of safety helmet missing.Aiming at this problem,It is recommended to use the safety helmet inspection method to combine the machine tool with multiple characteristics and auxiliary carriers In order to improve the recognition efficiency of safety helmet.Aiming at the problem that workers in thermal power plant wear clothes with similar color to safety helmet,and Meanshift algorithm(Meanshift)tracking will fail in the case of occlusion,a method combining kalman filter and multi-feature means shift algorithm is proposed to track the safety helmet detected.The main contents of this paper are as follows:Firstly,the VIBE algorithm is very dominant in detecting moving targets in terms of speed,and the speed of background modeling is fast.Therefore,the visual background difference algorithm(VIBE)is used in this paper to detect workers in the moving process,and then the initial location of the helmet area is carried out based on the proportion between the head and the whole body.Aiming at the problem that VIBE algorithm is prone to ghosting in the process of detecting moving targets,this paper detects and eliminates ghosting areas by calculating the cosine similarity and Euclidean distance of two-pixel distribution histogram between foreground target area and neighborhood background area.Secondly,due to the large area of the thermal power plant,the workers in the video vision are relatively small,and the helmet on the head is not easy to detect.In this respect,a method of helmet determination by multi-functional integration and vector support is proposed.First,extract the gradient distribution of image histogram(HOG)under test,optical flow histogram(HOF)two kinds of characteristics,because these two kinds of feature vector dimension is higher,the training time is longer,affect theefficiency of the helmet detection and,therefore,this paper USES principal component analysis(PCA)to dimension reduction of feature vector,to avoid the interference of redundant information.Then the two feature vectors after dimensionality reduction are cascined with the center of gravity and trained by SVM classifier to identify the safety helmet.Finally,the safety helmet in motion is tracked,aiming at the problem that the workers in thermal power plant wear clothes with similar color to the safety helmet,the video image resolution is low,and the Meanshift algorithm tracking will fail in the case of occlusion.In this paper,the Meanshift algorithm of multi-feature fusion is combined with kalman filter to study the safety helmet tracking method.First on the dress and helmet color interference,the target characteristic is not obvious,the Meanshift on the basis of the original color features,color information space will describe color histogram and describe the texture characteristics of LBP histogram statistics and combined with Meanshift algorithm,realized the dress color and the helmet helmet color close location accurate tracking.Aiming at the problem that the Meanshift algorithm cannot accurately detect the occluded safety helmet,this paper combines Kalman filter and Meanshift algorithm.The prediction function of Kalman filter is used to realize the accurate detection of shielding helmet.The experimental results show that the helmet detection method based on multi feature fusion and support vector machine(SVM),which is combined with the combination of multi-feature fusion and support vector machine(SVM),has better detection performance,and the average recognition rate of multiple groups of experiments is 90.03%.The helmet tracking algorithm,which combines Kalman filter and multi-feature fusion,can improve the tracking accuracy and reduce the number of iterations.