Design And Implementation Of Object Detection System Based On Scene Understanding

Object detection is one of the essential tasks in the field of computer vision.Its purpose is to accurately and efficiently find out the object concerned in the image,and to detect the category and location information of the object.In recent years,object detection algorithms based on deep learning have developed rapidly.However,most algorithms tend to focus only on the regional feature information of each object in the image,ignoring the influence on the result,caused by the information of the scene in which the object is located,such as the scene category,the connection between objects and other factors on the detection result,thus the utilization rate of image information is low.Therefore,the study of object detection method based on scene understanding is conducted in this thesis.The main research contents of this thesis are as follows:A scene classification method based on multi-scale feature fusion is proposed.According to the feature types,different network models are selected to extract global features and semantic features respectively,and the algorithm can gain more information in the image by integrating the two features.In order to solve the problem of deal with the imbalance of the number of parameters between the two features,an attention mechanism is introduced in the feature fusion stage to make the algorithm focus on important information.Experiments show that the accuracy of this algorithm is improved compared with common image classification networks on the scene15 data set and MIT Indoor data set.An object detection method based on scene understanding is proposed.The object relation calculation module is added into the object detection algorithm,and the relation between objects in the image is processed by a graph convolution neural network.At the same time,the calculation module of the relation between scene and object is added,and the relation between scene and object is processed by the circulating neural network.The method of bounding box regression is improved to make the final prediction more accurate.Based on the YOLOv5 s model,according to the comparison between the proposed method and existing target detection networks,the model structure proposed in this thesis has relatively good target detection results.The object detection system based on B/S architecture is designed and implemented.The system requirements,functional design and system implementation are planned in detail,and the related functions of the system are realized.