The Object Detection Algorithm Based On Context

In recent years,with the popularity of the Internet as well as the rise of video sites and social networks,people have access to a large number of multimedia resources such as images and video.As a result,computer vision has developed rapidly,and object detection has received more and more attention.As a classification problem,object detection plays an important role in the research and development of computer vision and machine learning.Object detection is widely used,such as face detection,pedestrian detection,vehicle detection and image classification.In order to achieve the goal of detection,object detection is usually divided into two sub tasks:object classification and object location.Object classification is to judge whether there is a class object that is detected in the image.If so,classify the object class according to the classification probability.The object location is to find the location of the object to be detected.Usually,it's a rectangular box.The traditional object detection algorithm is usually divided into three steps.The first step is to select a region with a sliding window.The second step is to extract the feature from this region.Finally,the regional feature is classified and the result is obtained.Such as face detection,first select the sliding window on the image,extract feature such as LBP or HOG,and then use SVM or AdaBoost classifier for classifi-cation to determine whether the current window is human face.From the 2006,deep learning method is gradually spread and has had a great influence on the field of com-puter vision.The object detection used deep learning developed fast.Deep learning method used' Region Proposal' select fewer windows and achieve higher recall.Deep learning method used regression greatly accelerate the speed of detection.When target detection is carried out,problems such as deformation,occlusion and visual angle change often exist,which results in poor detection results.Many studies have shown that the use of local context,global context and target context can reduce the impact of these problems and improve the detection accuracy.In order to address these issues,this paper proposes a traditional method and a deep learning method based on context to detect object.The main research achieve-ments are summarized as follows:1.Based on LBP,a new statistical method of feature histogram is designed,which adds local context information.There are two main changes.One is to expand the histogram statistical region,and the other is to give different weights to dif-ferent locations.2.Based on YOLOv2,a deep learning method for contextual target detection is designed,which adds target context information.First,the correlation between classes is calculated on the training dataset.Then,the YOLOv2 convolution net-work is used to obtain the bounding boxes and the classification probabilities of all classes.Class in bounding boxes with the highest confidence is selected as the reference class.The classification probabilities of all classes are changed accord-ing to the inter class correlation.Finally,the class with the highest probability of classification is used as the specified class.the probability whether the window contains the specified class object is calculated.The windows below a threshold are filtered.Finally,for the two methods mentioned above,experiments are carried out on ORL face dataset and PASCAL target detection dataset respectively.The experimental results show that the proposed method can achieve higher detection accuracy.