Fast Detection And Location Of Objects Based On Deep Learning Algorithm To Preferentially Capture

The core of this paper is to accurately and quickly realize the detection of chess in the game system.The traditional image processing algorithm is easy to be affected by the application environment because of the long processing pipeline,and the detection effect is poor in complex scenes.Lighting conditions,object size,orientation,etc.will affect the effectiveness of the detection algorithm?Therefore,traditional algorithms have higher requirements on the environment and require professional equipment to solve the environmental impact.Compared with the traditional image processing methods,the deep learning algorithm has lower requirements on light,object orientation and image quality,and can better adapt to various complex environments.This research detects chess based on deep learning algorithm.There are problems with small objects,small data differences,duplicate data,etc.in this task.These problems lead to complex algorithm structure and slow inference speed.In the actual chess process,it must be ensured that all objects are detected quickly and correctly.Because the difference in chess data is smaller,more differential data is needed to help neural network learn the characteristics of chess pieces.In view of the above problems,this paper optimizes the neural network algorithm and labels more differentiated data by using different optimization methods to meet the accuracy and speed requirements of the game system.This study built the hardware platform of the game system and labeled the object data set.The implementation includes Faster RCNN,SSD,YOLO algorithm,using different backbone networks to ensure that the algorithm has better detection effect and faster detection speed.By using OpenVINO and TensorRT to optimize algorithms for CPU and GPU,methods such as computational graph structure optimization,hardware device optimization,and numerical quantization are used to accelerate model reference.The final algorithm overcomes the shortcomings of traditional image processing algorithms for poor multi-object recognition and easy to be affected by light and object size,and can detect and move objects under extreme dark conditions.The algorithm can correctly detect the object of all the pieces in about 5ms.Finally,the detection of common item was tested under this platform.The test shows that the software and hardware platform can be applied to the capture of common objects,which has certain value in practical applications.