一种改进的DeepSORT矿井人员跟踪算法

An Improved DeepSORT Tracking Algorithm of Mine Pedestrian

煤矿井下光照不均、照度低且粉尘大,视频成像夹杂着噪声,进行视频监测时存在遮挡以及误检率高等问题。为保障井下人员安全,提出一种改进的DeepSORT目标跟踪算法,实现对矿井人员的跟踪。首先,选用OSNet全尺度网络优化浅层残差网络,提高表观特征提取能力;其次,优化交并比(Intersection over Union,Io U)匹配方式,采用完全交并比(Complete Intersection over Union,CIo U)匹配方式判断检测框与边界回归之间的匹配程度;最后,基于Python平台对改进后的跟踪算法进行仿真验证,检验算法的有效性。实验结果表明,发生遮挡时,与DeepSORT算法相比,改进算法增强了模型的健壮性,具有更好的跟踪效果。

In coal mine,the illumination is uneven,the illumination is low and the dust is large,the video imaging is mixed with noise,and there are problems of occlusion and high false detection rate in video monitoring.In order to ensure the safety of underground personnel,this paper proposes an improved DeepSORT target tracking algorithm to track mine personnel.Firstly,OSNet full-scale network is used to optimize the shallow residual network to improve the ability of apparent feature extraction.Secondly,the matching method of Intersection over Union(loU)was optimized,and Complete Intersection over Union(CloU)matching method was used to judge the matching degree between detection frame and boundary regression.Finally,based on Python platform,the improved tracking algorithm is simulated to verify the effectiveness of the improved algorithm.The experimental results show that compared with the DeepSORT algorithm,the improved algorithm enhances the robustness of the model and has better tracking effect when occlusibn occurs.