Research On Contextual Information Object Detection Algorithms Based On Deep Neural Network

With the rapid development of Internet technologies,the importance of computer vision is growing.O ne of the essential yet challenging problems in computer vision is object detection.In real life,because of changes of image conditions and various interference factors,the appearance of objects is significantly affected,resulting in differences between the objects.Even objects of the same category may affect the accuracy of object detection due to the interference factors such as imaging angle,imaging weather and light,object background,and imaging distance.This thesis completed the following task to improve the accuracy of object detection and reduce the amount of calculation:(1)Firstly,we introduce the research background and significance of object detection,then the composition of deep neural network and the process of object detection based on candidate frame and regression are briefly described.Many factors affect the effect of object detection,and this thesis adds contextual information as auxiliary information to address these issues in object detection.(2)This thesis proposes an object detection algorithm based on multi-layers context convolutional neural network and extracts contextual information of multiple layers to combine local features of objects in object detection.It also introduces the structure and characteristics of the experimental dataset,and it compares with the classical algorithm Faster R-CNN which does not add context information and some algorithms with adding context information.The detailed experimental results show that the proposed method is more accurate for object detection,especially for the detection of small objects.(3)The analysis of experimental results also shows that adding context information without distinction may not achieve the desired result.We hence propose a Convolutional Neural Network(SC D-CNN)target detection model with a differential context selection module that decides whether to add contextual information to assist object detection.After extracting candidate features of object and multi-region contextual information,calculating the difference between them,and comparing it with threshold,we can adaptively choose whether to integrate the contextual information.The experimental results demonstrate the effectiveness of the algorithm for object detection.