Design And Implementation Of Deep Learning Detection And Recognition System For Interactive Display

With the development of science and technology,the artistic form of incorporating scientific and technological elements is becoming more and more popular.Integrating deep learning and computer vision technology has great research potential and application value.Digital interactive display platform is mainly composed of target detection and recognition system and digital animation projection system,in which target detection and recognition system is the core part.Aiming at the problem of real-time recognition and location of object blocks in interactive display platform under complex environment,this paper proposes a fast recognition and location method based on YOLO V2 algorithm model.Combined with the size difference of object in sample image,a clustering algorithm is used to extract a priori frame scale that more matches the size of object in sample image,and a deep learning detection and recognition system for interactive display is designed and built.In order to ensure the complexity and diversity of the target sample image,data enhancement is used to expand the object block training sample.Finally,the performance of the system is verified by experimental analysis,and the end-to-end target block detection and recognition is realized.The main work of this paper includes:For this project,a specific project data set HUSTC605 is constructed.The sample images of the data set take full account of the external environmental factors such as the background,illumination intensity,illumination color and shooting angle of the target.The dataset is expanded by using data augmentation method.The data samples are randomly transformed into mirror images,interpolated by higher order,increased by Gaussian noise,salt and pepper noise and periodicity noise.Based on YOLO V2 model,a real-time object detection and recognition system is designed,and an object detection and recognition algorithm based on convolution neural network is developed.A real-time object detection and recognition system software is designed for testing the performance of the system based on Qt UI graphical interface.The software functional interface includes interface display control,camera calibration control,parameter setting control and other elements.The performance of system detection and recognition is validated at two levels: static test set and dynamic random real-time object block.The object blocks in several typical scenarios are tested and studied.The experimental results show that under six typical scenarios,the detection and recognition accuracy of the two levels tend to be 1;the recall rate is basically above 0.95,and the location accuracy of a single target is relatively high.In general,the detection and recognition system has excellent performance and robustness.