Recognition Traffic Police Actions By Bag Of Words Based On Hidden Markov Model

With the rise of autopilot technology,more and more attention has been paid to the related technologies.As one of the optical image processing technologies,motion recognition technology has become an indispensable technology in autopilot technology.The purpose of this paper is to use the relevant knowledge of color space to find the traffic police accurately in the complex environment,and to extract the action representation of the traffic police when conducting the traffic command,and then to express the traffic police action with the descriptor.Then using the bag of words to distinguish the action of the traffic police,through the hidden Markov model to identify the action posture and action sequence of the traffic police at this time,finally the line of sight to understand the command signal,so as to identify the command action of the traffic police.In order to better achieve the purpose of autopilot.The recognition process of traffic command action by hidden Markov model and trigonometric function is composed of four steps: searching for traffic police,extracting motion representation,describing action with visual word bag and judging command signal.The main work of this paper is as follows:The Retinex algorithm is improved,under the condition of maximum preserving the color of the image itself,the illumination is uniform,the error caused by the brightness problem is reduced,then the color that does not accord with the color of the traffic police is excluded by the color threshold,and then the region grows.Filtering,morphological processing to obtain the color color in accordance with the threshold value of the region contour,and then these contour map chromatographic distribution judgment,to find out the color of the uniform color of the traffic police chromatographic distribution area,so as to find the traffic police.By changing the traditional optical flow field and tracking the moving position of the traffic police through regional growth,the moving direction of the traffic police in two-dimensional space is determined,and the moving area is delineated.Then the invariant moment is used to find out the deformation of the moving region,and through the deformation produced by the moving region,whether the deformation occurs in the three-dimensional direction is determined.According to the recognition method of bag of words,the original motion image is described with various motion descriptors,and the motion vectors in different dimensions are compared with the motion region of the trunk.The movement of the former moment is combined into a visual word bag library,and then the traffic police's posture is judged by recognizing the characteristics of the observed traffic police action.Based on the hidden Markov model instead of the traditional Bayesian network,the action posture and action sequence are modeled,and then the action posture of the traffic police is recognized by the action character lexicon described above.Then the traffic police action sequence is judged by the posture characteristics of the traffic police,and finally the command signal of the traffic police is recognized by the line of sight.