Hand Gesture Recognition Based On Infrared Camera

With the continuous improvement of material life,researchers begin to explore human-computer interaction technology in order to conform human communication habits.Among them,vision-based gesture recognition technology can better meet the human-oriented design concept and provide necessary assistance for machines to understand human postures.Compared with traditional input methods,vision-based gesture recognition is easy to operate without direct contact.It has become a hotspot in human-computer interaction research,and has a very wide range of applications in the fields of sign language recognition,assisted driving,equipment operation and smart home.However,most of the current studies have been restricted to well-lit environments.For all-weather missions,especially in the military and security fields which need to be performed at night,ordinary color camera is obviously incapable,which greatly restricts practical process.Out of the above-mentioned situation,it proposes a new method which combines gesture recognition technology with infrared night vision technology to recognize gesture under low illuminance.Specific work is as follows:It uses active infrared technology at close range to capture infrared images,but the quality of the images taken at night is poor and the noise interference is serious.Then it discusses the main causes of noise,and analyzes the characteristics of infrared noise through non-uniformity experiment.Aiming at the fringe presented by image noise,it proposes a correction method based on Kalman filtering,which removes wild fluctuations while retaining the completeness of the details and has practical effect on improving the quality of infrared images.According to the characteristics of infrared image,the great gray difference between hand and background can be used by segmentation algorithm based on K-Means clustering to extract gesture area and then explains the gesture classification rules based on geometric information.However,it is found that clustering algorithm cannot guarantee the completeness of gesture segmentation due to the uneven illumination of the light source.In addition,K-Means clustering algorithm is not stable enough,and can be easily affected by K value and initial seed point.To solve the problem that infrared image with uneven grayscale cannot obtain effective features,it proposes classification algorithm which combines HOG with SVM.By establishing feature model without relying on image segmentation,experimental results show that the algorithm has certain enhancement in the precision of recognition.By correction method,infrared noises achieve elimination and the peak signal-to-noise of image improves in a certain extent.Above all,assessment results show that the accuracy of gesture recognition is up to 91.8%.In conclusion,the algorithm can meet the goal of recognizing different gestures in low-illumination environment.