The Improvement And Application Of DLMA Skeleton Extraction Algorithm

The discrete lambda-medial axis（DLMA）is a fast and robust medial axis transformation.It can be applied in extracting single-pixel accurate skeletons.But an appropriate parameterλis needed to set in advance.Meanwhile,it relies on the single threshold filters.Thus,it is hard to select the highly adaptive parameterλ,when deal with the shapes having complex topologies.We proposed a method combined the local maxima of Euclidean distance transform and the idea of background point space.The proposed algorithm divides the DLMA algorithm into two steps:Firstly,a smallλthreshold is used to obtain the rough extraction result of the skeleton,meanwhile,the N₈ neighborhood is reduced to N₂ neighborhood;Secondly,The adaptive skeleton growth strategy is used to obtain the accurate skeleton under the rough extraction result.The experimental results show that the proposed optimized algorithm has stronger adaptive ability,stronger robustness and faster computing speed,which is suitable for human behavior recognition neighborhood.Make full use of the advantages of improved DLMA algorithm,this paper established a gesture recognition system based on skeleton geometry model,and implementation in hardware.The system is constituted by general camera and ARM.Firstly,the dynamic foreground detection algorithm combining YCbCr color recognition model is used to segment the gesture area;secondly,by Euclidean distance transformation and improvement of DLMA European skeleton,we implement a geometry model of gesture recognition.The correct recognition rate is as high as 94%,each frame image processing time is less than 25ms,which shows that the system is real-time and effective.