Human Pose Estimation and Action Recognition System on Reconfigurable Dnn(2/3)

The technology and application of Artificial intelligence is become more and more common in our lives, whether it is the speech recognition, face recognition or object classification applications have a lot of research. The inability of traditional algorithms to extend to complex environments has been solve by deep neural networks via cuda cores accelerating. AI technology is superior to human on decision-making or judgment in many applications, therefore, AI has become the hottest research area. With more and more research that can be applied to real life published, the smart home is no longer a dream. We can imagine life in the future, appliances in house can be controlled by voice, and human behavior can be detected to automatically adjust light levels, air-conditioning temperatures, or tv volume, making life more convenient.Smart home is the main trend in the future, in this project, low-cost color image cameras will be used instead of expensive infrared cameras (Kinect, Realsense, etc.). The system is expected to achieve real-time processing and is similar to the real-life environment, for ordinary users, there will be a convenient and comfortable operating environment and no price burden. In this project, we will discuss how to use two image cameras to generate the depth images in complex environments without wearable devices requiring or pre-setting system. And explore how to use stereo vision technology to transform the dual camera stream into representative body parts features, which other researches can use to action recognition, health care or smart home application.