# NUUEE Computer vision Slides : https://hackmd.io/@shawnlintw/B1qHzbjO8 Report : Shawn Shih Hsiung Lin Student ID : M0822002 E-mail: edwin2619@gmail.com Date : 2020-04-21 --- <h2> Gesture Recognition demo </h2> - Live Demo ? Die ? --- <h2> How it works ? </h2> ```flow st=>start: Input Frame op1=>operation: Hand Image segmentation op2=>operation: Feature Extraction op3=>operation: Computer the Histogram of BOW. op4=>operation: Input the features into SVM. e=>end: Output result st->op1->op2->op3->op4->e ``` --- <h2> Skin segmentation </h2> * In this project, We choice using the hybrid mask that mixed HSV-color space and YCbCr-color space.  --- <h2> BOW : Bag of words </h2> ICCV 2009 Recognizing and Learning Object Categories  --- <h2> Generate the 'Word feature' </h2> 1. Feature detection (Here we use SURF.) 2. Codeword dictionary format by KMeans.  4. Re-mapping the features you extract from training samples to words that generated in step2. 5. You can get the Hitogram of the 'word features' for each image. --- <h2> Training SVM </h2> ```flow st=>start: Input Frame op1=>operation: Hand Image segmentation op2=>operation: Feature Extract "ROI" from all category of training set op3=>operation: Generate the word features by KMeans op4=>operation: Feature Extract "ROI" and re-mapping these feature to word feature. op5=>operation: Use the word feature to training the SVM e=>end: Output result st->op1->op2->op3->op4->op5->e ``` --- - For this procedure, we found the success rate of recognition is dependent on skin segmentation.     --- - Coding step : https://hackmd.io/@shawnlintw/By0_Fac_8 - Source Code : (Open source by GPLv3 ) https://github.com/shawnlintw/Gesture_recognition ---