Hand Detection And Tracking For Gesture Recognition

Thanks to its convenient and natural,the new human-computer interaction technology of gesture recognition is increasingly used in virtual reality,rehabilitation,remote control and other fields.Hand detection and tracking are two basic and core issues behind gesture recognition.Its accuracy and real-time directly determine the user experience and efficiency of gesture human-computer interaction technology.Therefore,this thesis focuses on the hand detection and tracking methods,aiming to provide efficient and accurate hand detection and tracking algorithms for gesture recognition human-computer interaction.Aiming at the difficulty of the hand detection technology,there is a huge difference in the data class.This paper proposes a hand detection method based on conditional random forests.First,cluster analysis is performed on the human hand.Since the hands with similar postures and illumination condition are clustered into the same class,the differences in the hand types in each class after clustering will be significantly reduced.Secondly,conditional random forests are constructed for the human hands in each cluster.Finally,according to the distance between the test sample and each cluster center in the feature space,a corresponding number of decision trees are selected from the corresponding conditional random forests,and a random forest is dynamically constructed to test the classification of the samples.Experiments show that the hand detection method proposed in this paper has good accuracy.Aiming at the problem that the tracking algorithm is easy to be affected by background interference,this paper proposes a hand tracking method based on the discriminative target model with color features.On the one hand,the minimum error rate Bayesian classifier based on color histogram is established to separate the human hand and the background;on the other hand,the interference region similar to the appearance characteristics of hand is identified and modeled into the Bayesian classifier,further Suppress background interference.In addition,a segmentation threshold adaptive method is proposed to more accurately segment the hand and the background.Experiments on challenging datasets show that the proposed hand tracking algorithm can reduce the interference effects in the background area,thus reducing the occurrence of drift phenomena,with good accuracy,robustness and real-time.