Research And Application Object Detection Algorithm Based On SSD

Object detection is one of the core foundations of computer vision and has very important research and application value.The wave of deep learning in recent years has effectively promoted the rapid progress of object detection.The progress of object detection has also led to the development of the entire computer vision field.Nowadays,object detection has been widely used in many fields,such as: automatic driving,intelligent monitoring,robot intelligence,etc.Therefore,how to design a more balanced and reliable object detection network is an urgent problem in academic and practical application fields.In response to academic and practical application requirements,this paper is based on the field of universal object detection,focusing on how to improve the performance of existing object detection algorithms for a series of research.The main contributions of this paper can be summarized as follows:(1)Aiming at the problem of slow detection speed of object detection algorithm,this paper proposes the fast SSD detection algorithm,taking SSD as the basic network framework,introducing the idea of depthwise separable to build a high-speed reasoning framework for the object detection network,which will effectively reduce the complexity of the network and improve the speed of detection.(2)Aiming at the problem of reduced accuracy brought by the reduction of feature correlation caused by the depthwise separable construction of the network,this paper proposes the best balanced and reliable SSD object detection model(BBR-SSD).It first builds a Feature Pyramid Networks(FPN)on top of the fast SSD,then,on this basis,the Adaptive Spatial Feature Fusion(ASFF)unit is further combined.The organic combination of FPN and ASFF can effectively fuse shallow features and deep features information.It can improve the accuracy of object detection while maintaining a low amount of parameters and calculations.(3)Aiming at the problem of missed detection when the current object detection post-processing step filters out duplicate boxes,this paper proposes to introduce the Soft Non-Maximum Suppression(Soft-NMS)algorithm to replace the traditional NonMaximum Suppression(NMS)algorithm to filter the detection box.Under the premise of controlling the calculation cost,it can effectively solve the missed detection situation.In this paper,the proposed algorithm is tested on public data set,and the results show that the algorithm can effectively improve the accuracy of the one-stage object detection while maintaining a high inference speed.In order to test the balance and reliability of the algorithm in this paper,the algorithm is deployed in specific scenarios,and corresponding experiments are carried out for the practical application fields of pedestrian detection and face detection.Finally,the results of practical application in specific scenarios also show the robustness of the proposed algorithm.