A Small Moving Object Detection algorithm Based on Motion Information (SMOD-BMI) is proposed to detect small moving objects with a low Signal-to-Noise Ratio (SNR) in surveillance video. Firstly, a ConvLSTM-PAN model structure is designed to capture suspicious small moving objects, in which the Convolutional Long and Short Time Memory (ConvLSTM) network aggregated the Spatio-temporal features of the adjacent multi-frame small moving object and the Path Aggregation Network (PAN) located the suspicious small moving objects. Then, an object tracking algorithm is used to track suspicious small objects and calculate their Motion Range (MR). At the same time, the size of the MR of the suspicious small moving object is adjusted adaptively according to its speed of movement (specifically, if the object moves slowly, its MR will be expanded according to the speed of the object to ensure the necessary environmental information of the object). Adaptive Spatio-temporal Cubes (ASt-Cubes) of the small moving objects are generated to ensure that the SNR of the moving objects is improved, and the necessary environmental information is retained adaptively. Finally, a LightWeight U-Shape Net (LW-USN) based on ASt-Cubes is designed to detect small moving objects, which rejects false detections and returns the position of small moving objects. This paper uses the bird in the surveillance video as the experimental data set to verify the algorithm's performance. The experimental results show that the proposed small moving object detection method based on motion information in surveillance video can effectively reduce the missed and false detection rate of small moving objects.

翻译：提出了一种基于运动信息的小运动目标检测算法（SMOD-BMI），用于检测监控视频中低信噪比的小运动目标。首先，设计了ConvLSTM-PAN模型结构以捕获疑似小运动目标，其中卷积长短期记忆网络（ConvLSTM）聚合了相邻多帧小运动目标的时空特征，路径聚合网络（PAN）定位了疑似小运动目标。随后，采用目标跟踪算法对疑似小目标进行跟踪，并计算其运动范围（MR）。同时，根据可疑小运动目标的运动速度自适应调整其MR的大小（具体而言，若目标运动缓慢，则根据目标速度扩展其MR，以确保获取必要的环境信息）。生成小运动目标的自适应时空立方体（ASt-Cubes），以提高运动目标的信噪比，并自适应保留必要的环境信息。最后，基于ASt-Cubes设计了轻量级U形网络（LW-USN）用于检测小运动目标，该网络剔除虚警并返回小运动目标的位置。本文以监控视频中的鸟类作为实验数据集验证算法性能。实验结果表明，所提出的基于运动信息的监控视频小运动目标检测方法能有效降低小运动目标的漏检率和虚警率。