基于深度学习的行人检测跟踪及距离测定研究

Research on Pedestrian Detection,Tracking and Distance Measurement Based on Deep Learning

突发的传染性疾病引起了广泛的关注,预防感染最佳的方法之一就是尽可能的避免接触到传染性疾病病毒,最直接的方式就是控制好距离,即保持人们之间的社交距离。基于深度学习提出了一种对于人群目标检测跟踪以及距离测定的方法。基于深度学习的算法有准确度高、实时性强、泛化能力强、有一定的抗干扰能力、良好的扩展性等优点。由于摄像机与地面有一定的倾斜角,而不是直接垂直朝下(正投影),所以文本采用透视变换算法将图像校正成正投影的形式,以便于我们确定目标之间的距离。采用Ghostnet轻量化主干网络作为Backbone网络的基础结构,将网络结构精简,减少了模型的计算量以及参数量,但同时也保证了检测的性能。

Characterized The sudden outbreak of infectious diseases has attracted widespread attention.Onc of the best ways to prevent infection is to avoid contact with infectious disease viruses as much as possible.The most direct way is to control the distance,that is,to maintain social distance betwccn pcople.Based on decp learning,a method for crowd target detcction and tracking as well as distance measurement is proposed.Deep learning-based algorithms have the advantages of high accuracy,strong real-time performance,strong generalization ability,certain anti-interference ability,and good scalability.Due to the camera's inclination angle relative to the ground,rather than pointing directly downward(orthographic projcction),the text uses a perspective transformation algorithm to corrcct the image into an orthographic projection form,allowing us to determine the distance between objects.Ghostnet lightweight backbone network is used as the basic structure of the Backbone network,which simplifies the network structure,reduces the computational load and parameter quantity of the model,but also ensures the performance of detection.