Research And Application On Hand Detection Algorithm For Mobile Terminal With Complex Background

With the development and integration of mobile devices and human-computer interaction technology,the research of the mobile terminal detection technology is becoming more and more important.Hand detection for mobile terminal with complex background faces two challenges: the limitations of computing resources and storage space of mobile devices and the influence of complex background.Therefore,the algorithm should have high recognition accuracy and computational performance.To handle the problem of hand detection for the mobile terminal with complex background,this paper considers it from two levels: pixel-level and region-level.For pixel-level detection,we propose a new method called variable features based structured forest;for region-level detection,we propose an improved version ACF algorithm.The main work:1.Analyzing the challenges of hand detection for mobile terminal with complex background in detail and providing the assumptions and solutions relatively.In addition to the limitations of the computing resources and storage space of mobile device,challenges mainly involves five aspects: 1)the perspective of three-dimensional space in different 2D images will cause dramatic changes;2)the changes in light conditions;3)the diversity of hand gestures;4)containing color object background;5)non-integrity of hand contour.2.Aiming at the problem of pixel-level hand detection,this paper introduces variable feature and structured learning into the framework of random forest.In this algorithm,the nodes would be split in different feature spaces,and the structured output space is used to enhance the discriminative and generalization ability of the algorithm.The result of pixel-level hand detection would distinguish the pixels belong to hands in the image.3.In order to detect hands at regional-level,this paper proposed an improved ACF algorithm based on multi-color spaces,structured edge detection,Edge Boxes,and Spatial Pyramid pooling.Then it would be combined with AdaBoost for hand detection.It would mark out the rectangular area containing a hand.4.Finally,this paper takes the mobile terminal gesture drawing system as the prototype system,and uses the above method to detect the hand,get the hand trajectory,and complete the operation of the moving graphic object.