Parallel Gesture Recognition System Based On Deep Learning In Complex Background

With the popularity of mobile terminals and wearable devices, people want to replace the current methods used during the interacting between human and computers with a more natural way, and gesture recognition based on machine vision is the one we are looking for.In the context of human-computer interaction based on computer vision, we use deep learning algorithms to do gesture recognition research,and try to build a complete and usable system using embedded technologies and parallel computing technologies. To get the actually usable natural gesture images, this paper use ARM + Linux + USB camera to capture natural color gesture images under the indoor complex background. Affected by the cloud computing, we migrate image processing to a PC platform because of its more efficient in computing and storage;taking into account the growing network security problems and constraints of the mobile terminal battery problems, we use moving object detection, H.264, Open SSL to compress and encrypt data during transmission, saving resources and ensuring data safety meaningful.After the PC receives the data, we begin picking up the gestures. To make the system, whose recognition time should be as short as possible, actually available and upgradable in intelligent, we use deep learning algorithms to recognize gesture because its test time is very short. Then we use a single GPU to accelerate the training and the recognition of deep network.Deep Learning algorithms wouldn’t be so good unless there is adequate training samples, we use de-noise auto-encode algorithm to improve the model performance slightly, and verify the self-learning algorithms on a small amount of samples.