The Design And Implementation Of Augment Reality System Based On SSD Object Detection Algorithm

Augmented Reality(AR)is the technology that fuses real world with virtual information.As the key technology in smart device application,AR has huge influence in enriching interactive mode and improving user experience.Object detection technology is important for AR system.Traditional way cannot detect target object accurately in complex environment and cannot estimate the position of target object.With the development of object detection algorithm based on deep learning,the model trained by big data can detect the natural object in the environment with high accuracy and speed.However,this kind of model is complex because the parameters are float real-valued,which brings high computational complexity and occupies large memory size.It's hard to apply on low-power devices.(1)Combine SSD with Binary Neural Network,introduce a method to train SSD with binary weights and activations,it can run faster and reduce memory size without suffering any loss in classification accuracy.(2)To estimate the target object accurately,this paper uses a pose estimation algorithm based on target object,transforming the coordinates of target objects in image to the 3D position in the new world.The experiment verifies that the method meets the requirement of the system's accuracy.(3)Apply AR technology to cultural heritage to develop a museum tourism system.The system can recognize the natural object in the environment,display 3D content about antiques such as models,labels,videos.