Research And Application Of Object Detection Algorithm Combined With Super-resolution Technology

Object detection is one of the research hotspots in the field of computer vision,and its research achievements are widely used in intelligent security,autonomous driving and other fields.With the development of deep learning technology,object detection technology has made a breakthrough development,but it still has great limitations in practical applications due to the lack of accuracy for small object detection and low-resolution object detection.So,it has a great significance for promoting the application of object detection that how to develop an algorithm can accurately detect small objects and low-resolution objects.Based on the problem that the object detection technology is not accurate enough for small objects and low-resolution objects,it is proposed to apply super-resolution technology to object detection after researching and breaking through related technologies.The main research contents include:(1)Applying subpixel shuffle technology to FSSD network,redesigned FSSD's feature fusion module to strengthen its ability for feature map information of different size,and named the network using the new feature fusion module as SFSSD.The mean average precision of SFSSD on Pascal VOC 2007 test set is 78.94.(2)Based on SFSSD,a multitask object detection algorithm which combines the superresolution network(SRCNN)is proposed,and it is named Super-Resolution Single Shot Multibox Detector(SRSSD).SRSSD can deal with super-resolution reconstruction and object detection of input images in a single network.The result of super-resolution reconstruction is used as the input of the object detection sub-network to improve the detection performance of the network.The detection performance of small objects and lowresolution objects improve obviously.The mean average precision of SRSSD on Pascal VOC 2007 test set is 79.40.(3)The intelligent perimeter warning system based on SRSSD is realized.It uses SRSSD and KM algorithm as the core algorithm,which can realize the automatic detection and giving alarms to border crossing,intrusion and others.The detection performance of this system which using SRSSD is better the one which using SSD in terms of small size pedestrian,moving and blur images.Research on how to detect small objects and low-resolution objects accurately can make the application of object detection technology more extensive,and it has strong theoretical and practical value.