Traffic Police Gesture Recognition Based On Depth Image

With the arrival of the age of intelligent transportation,unmanned driving has become the mainstream trend of development.The study of human-computer interaction with hand gestures will be of great significance.The research of human-computer interaction on traffic police gestures will be very important significance and value.With the help of traffic police gestures,it can not only make friendly interaction with machine,but also can achieve emergency treatment in the congestion,accident and signal failure of traffic.From the perspective of intelligent traffic and human-computer interaction,this investigation combines the Kinect sensor with deep information acquisition equipment.Aiming at the limitation of hand gesture recognition at present,this research mainly studies the traffic police gesture recognition based on depth image,and improves the feature extraction method in the recognition process,and the better recognition results are realized.The specific research content of this paper is as follows:Aiming at the problem that the traffic police gesture recognition based on two-dimensional vision is limited by light and occlusion,and the limitation of fixed position of traffic police gesture recognition based on skeleton information,a method of traffic police gesture recognition based on depth image is proposed.This method combines depth image with gesture recognition method based on two-dimensional vision: firstly acquires the key traffic police gestures in the depth video frame sequence,and preprocesses traffic police gestures;then,combines the thinning algorithm and Radon transform to extract features of traffic police gestures;Finally,using the DTW algorithm to achieve dynamic traffic police gesture recognition.The results show that this method can achieve better recognition of traffic police gestures under different lighting,distance and location,and overcome the shortcomings of current traffic police gesture recognition.The EPTA thinning algorithm is easy to produce branching problems after the image of the unsmooth contour is refined,and the Radon transform is sensitive to the distribution of pixels in the thinning line.An improved EPTA thinning algorithm and a re-evaluation of the Radon transform curve are proposed.The thinning algorithm is improved by a deleting restriction and iterative scaling mechanism.First,it regularly refines the pixels that do not produce branching;then,it performs global smoothing and refinement to solve the branching problem caused by some pixels.The Radon transform re-estimation re-estimates the Radon transform curves at different angles of similar thinning lines by means of neighborhood averaging.The experimental results show that the improved thinning algorithm can obtain the results without branches and smooth refinement;Radon transform re-estimation reduces the sensitivity of Radon transform to thinning lines and reduces the difference in transform curves.The improved feature extraction method can describe the feature information better and improve the performance of traffic police gestures.Although this investigation based on the depth image to achieve the recognition of seven kinds of traffic police gestures,taking into account the actual application of traffic police gestures in the complex environment,the real-time traffic police gestures interaction and traffic police gestures are not fully recognized and so on,we still need further exploration and research.