Research On Moving Object Detection Algorithm Based On Superpixel And Deep Learning

Moving objects are one of the important research objects in the field of computer vision.The moving object detection algorithm can effectively extract moving objects of interest from the video.However,challenges such as camera shake,dynamic background,and ghosting effects in real-world situations can significantly reduce the accuracy of moving object detection.In view of the above problems,we have carried out the following two aspects of work.(1)A superpixel-based approach for moving object detection is proposed.First,the superpixels of the first frame are constructed using a simple linear iterative clustering method and extract the standard deviation features of each superpixel to enhance the anti-noise capability.Then,the initial superpixels are divided into K smaller sub-superpixels via the k-means clustering algorithm,and background model is then initialized by representing each sub-superpixel as a multidimensional feature vector.For the subsequent frames,moving objects are detected by the weighting measure.In order not to lose the small object,the weighting measure takes into account the change in the number of pre-pixels of the superpixel,at the same time,each superpixel is filtered twice in the detection process to reduce the number of calculations and improve the detection speed.Finally,in order to deal with ghost artifacts,a background model updating strategy is devised,based on the number of sub-pixel members per sub-pixel.Experimental results show that compared with the known methods,our algorithm achieves a competitive accuracy for background subtraction while keeping less running time,it is robust to noise and not easy to lose small objects while can quickly solve the ghost effect.(2)A moving object detection method based on U-Net network of deep learning is proposed.First,the U-Net convolutional neural network is constructed to reduce the dependence of deep learning on the number of training data sets,so that the algorithm can be an excellent model trained on a small data set.Then,the ratio of the number of positive and negative samples in the data set is calculated,and the reciprocal of the ratio is used as the sample weight to cope with the positive and negative sample imbalance.Finally,a threshold is set to threshold the prediction result to obtain a moving object detection result.Experimental results show that the algorithm requires less frame images as the training set to obtain extremely high detection accuracy,and the detection result is not significantly biased moving objects or background.