Moving Object Detection In Video Research Articles

Highly Efficient and Unsupervised Framework for Moving Object Detection in Satellite Videos.

Moving object detection in satellite videos (SVMOD) is a challenging task due to the extremely dim and small target characteristics. Current learning-based methods extract spatio-temporal information from multi-frame dense representation with labor-intensive manual labels to tackle SVMOD, which needs high annotation costs and contains tremendous computational redundancy due to the severe imbalance between foreground and background regions. In this paper, we propose a highly efficient unsupervised framework for SVMOD. Specifically, we propose a generic unsupervised framework for SVMOD, in which pseudo labels generated by a traditional method can evolve with the training process to promote detection performance. Furthermore, we propose a highly efficient and effective sparse convolutional anchor-free detection network by sampling the dense multi-frame image form into a sparse spatio-temporal point cloud representation and skipping the redundant computation on background regions. Coping these two designs, we can achieve both high efficiency (label and computation efficiency) and effectiveness. Extensive experiments demonstrate that our method can not only process 98.8 frames per second on 1024 ×1024 images but also achieve state-of-the-art performance.

DSFNet: Dynamic and Static Fusion Network for Moving Object Detection in Satellite Videos

Moving object detection (MOD) in satellite videos remains challenging due to the extremely small size of the interested targets and the highly complex background. Both the intra-frame (static) and inter-frame (dynamic) information are of great importance to MOD. In this letter, we propose a two-stream detection network named dynamic and static fusion network (DSFNet) to tackle the MOD problem in satellite videos. Specifically, the DSFNet is composed of a 2-D backbone to extract static context information from a single frame and a lightweight 3-D backbone to extract dynamic motion cues from consecutive frames. Then the extracted static and dynamic features are fused and fed into the detection head to detect the moving targets in satellite videos. We conduct extensive experiments on videos collected from Jilin-1 satellite and the results have demonstrated the effectiveness and robustness of the proposed DSFNet. Experimental results show that our DSFNet achieves the-state-of-the-art performance.

Robust Privacy-Preserving Motion Detection and Object Tracking in Encrypted Streaming Video

Video privacy leakage is becoming an increasingly severe public problem, especially in cloud-based video surveillance systems. It leads to the new need for secure cloud-based video applications, where the video is encrypted for privacy protection. Despite some methods that have been proposed for encrypted video moving object detection and tracking, none has robust performance against complex and dynamic scenes. In this paper, we propose an efficient and robust privacy-preserving motion detection and multiple object tracking scheme for encrypted surveillance video bitstreams. By analyzing the properties of the video codec and format-compliant encryption schemes, we propose a new compressed-domain feature to capture motion information in complex surveillance scenarios. Based on this feature, we design an adaptive clustering algorithm for moving object segmentation with an accuracy of 4×4 pixels. We then propose a multiple object tracking scheme that uses Kalman filter estimation and adaptive measurement refinement. The proposed scheme does not require video decryption or full decompression and has a very low computation load. The experimental results demonstrate that our scheme achieves the best detection and tracking performance compared with existing works in the encrypted and compressed domain. Our scheme can be effectively used in complex surveillance scenarios with different challenges, such as camera movement/jitter, dynamic background, and shadows.

Improved Algorithm of the Video Moving Object Detection Based on ViBE

Background: ViBE (Visual Background Extractor) is an algorithm with a variety of advantages in video moving object detection which utilizes a pixel-level background modeling. However, it is not suitable for distinguishing the scene of drastic change, adapts poorly to the sudden change of the illumination and may lost the object easily, because this algorithm uses a fixed threshold to distinguish the background from the foreground. Methods: In this paper, an improved ViBE algorithm is proposed, which an adaptive dynamic threshold method is introduced for classification of the foreground and the background in the changing scenes. When reconstructing the model it required for drastic change of illumination, Otsu algorithm is used to judge the threshold and select the appropriate frame to complete the reconstruction to achieve quick adapt to the light. Results: Experimental results show that the proposed algorithm has higher recall value, better precision and F value comparing to the original algorithm. The improved algorithm has the highest classification accuracy among other similar algorithms and therefore the improved algorithm significantly improves the detection results. Conclusion: After analyzing the principle of ViBE algorithm, this paper proposed improvements to it from two aspects to aim at its deficiency. Taking into account of the dynamic changes of different environments, the change factor was proposed to measure the dynamic degree of background. According to the value of the factor, adaptive clustering was obtained and clustering threshold was updated to improve the adaptability of the algorithm to the dynamic environments. The improved ViBE algorithm can find the appropriate frame to reconstruct model structure in the case of abrupt light change, which can quickly adapt itself to the light change and be more accurate in the object detection.

A Novel Method for Video Moving Object Detection Using Improved Independent Component Analysis

Moving object detection is a hot and difficult problem. How to separate the changed part from the unchanged one is the key. Also, the connection between the multi-dimensional information of input frames often plays an important role in moving object detection. In this article, we presented a novel and effective video moving object detection method. The innovation of this method lies in the calculation of the connection between multi-dimensional information considering time and space domain at the same time. At present, most moving object detection algorithms consider the time dimension and the space dimension in an independent manner. It is possible to cause some detection problems, such as, if the background information of a static or low-speed object is updated too quickly, only part of the foreground region is detected correctly. To avoid such deficiencies, the improved ICA method is adopted to separate the continuous multi-frame images in the time domain, and then to integrate these signals in the space domain. It could separate and detect the difference between the input signals to extract the correct foreground target, especially in some situations that insignificant changes happen. The proposed algorithm is named VTD-FastICA. A large number of experiments have shown that our method is more competitive than most of the moving object detection methods, especially in detecting long-span, occlusion environments, and objects with weak changes. Moreover, we also analyze the efficiency of our method in different situations, indicating that the proposed algorithm can be applied to real situations.

MSFgNet: A Novel Compact End-to-End Deep Network for Moving Object Detection

Moving object detection (MOD) in videos is a challenging task. Estimation of accurate background is the key to extracting the foreground from video frames. In this paper, we have proposed a novel compact end-to-end convolutional neural network architecture, motion saliency foreground network (MSFgNet), to estimate the background and to extract the foreground from video frames. Initially, the long streaming video is divided into a number of small video streams (SVS). The proposed network takes the SVS as an input and estimates the background frame for each SVS. Second, the saliency map is extracted using the current video frame and estimated background. Furthermore, a compact encoder-decoder network is proposed to extract the foreground from the estimated saliency maps. The performance of the proposed MSFgNet is tested on three benchmark datasets (CDnet-2014, LASIESTA, and PTIS) for MOD. The computational complexity (handling of number of parameters and execution time) and the performance of the proposed MSFgNet are compared with the existing state-of-the-art methods for MOD in terms of precision, recall, and F-measure. Performance analysis shows that the proposed network is very compact and outperforms the existing state-of-the-art methods for MOD in videos.

Encoder combined video moving object detection

Moving object detection is the initial step of information extraction for video surveillance applications. For efficient transmission and storage, surveillance videos are compressed after capture. Existing methods usually take the compressed videos for moving object detection, which need full or partial decoding. Thus, they are not very efficient. Actually, the data generated during video encoding can be shared with detection, which avoids extra decoding. In this paper, we propose a low computation moving object detection method combined with video encoder. To avoid extra decoding, data generated during video encoding, such as Rate Distortion Cost (RDCost), are utilized for moving object detection. Experiments on a number of video sequences demonstrate its effectiveness in detection accuracy and superior processing speed.

Research and Restoration Technology of Video Motion Target Detection Based on Kernel Method

Abstract In recent years, due to the video surveillance applications more and more widely, people are not satisfied with the goal of monitoring, and the video monitoring technology of intelligent video moving object detection and tracking technology has received extensive attention. The research work in this paper is in the field, the moving target detection spatiotemporal correlation and difference contour tracking algorithm based on a fixed background. The algorithm in the background under the condition of fixed to pay a smaller time complexity, the target detection and tracking has a good effect, so it has higher application value. In this paper, the prospect of caused motion detection of occlusion background foreground correlation problem, put forward the video moving object detection method based on kernel independent component analysis, canonical correlation to minimize the component in the high dimensional feature space in order to separate the foreground nuclear background. Independent component analysis assumes that the foreground and background independent, avoid the correlation problem. The two objective functions based on Kernel Independent Component Analysis: analysis of kernel independent component analysis based on kernel canonical component (KCCA) and kernel independent component analysis (KGV) based on the generalized variance. KCCA is the application of canonical correlation analysis in the kernel method, discuss is the first canonical correlation separation component of high dimensional map, and KGV are typical correlation between the components in the high dimensional space of the whole spectrum. Both KCCA and KGV improved the accuracy of motion detection.

An adaptive graph cut algorithm for video moving objects detection

The algorithms based on graph cut have the advantage to detect the moving objects effectively and robustly. The main trouble of the algorithm based on graph cut is that its model parameters will be determined empirically. In this paper, a novel algorithm of adaptive graph cut is proposed to detect video moving objects. Based on Markov random field model, the proposed algorithm uses the numbers of moving objects pixels and objectives-background pixel-pairs to describe the geometric features of the moving objects. And the relationship between the geometric features of the moving objects and the model parameters are set up. In this paper, the model parameters are adaptively optimized through the extraction and prediction of the geometric features of moving objects. Then the detection based on the graph cut is preformed on ROI, which well achieves the balance between the computation and accuracy. Finally, the experimental results show the proposed algorithm can hold the details of moving objects more effectively compared with other algorithms, and improve the detection performance of moving object in the video surveillance.

Steering kernel-based video moving objects detection with local background texture dictionaries

Texture and color information as important characters of video objects have been widely used in the detection of moving objects. A detection algorithm based on texture may cause detection errors in regions of blank texture and heterogeneous texture, and a detection algorithm based on color is easily influenced by illumination changes and shadows. In this paper, a new detection and fusion algorithm is proposed. At the detection stage based on texture, the background texture is classified according to the steering kernel. At the fusion stage, for the moving objects detected on the basis of texture and color respectively, a scheme based on a boundary selection strategy is proposed for combining the different detection objects. A relatively smooth boundary is selected as the true boundary, and the shadow detection is carried out to assist the boundary selection. Experimental results verify the advantages of the proposed algorithm as compared to the existing state-of-the-art algorithms.

Adaptive graph-cuts algorithm based on higher-order MRF for video moving object detection

A new graph-cuts algorithm for video moving object detection is presented. Using the area, perimeter and shape of the moving objects, the energy function from the low-order Markov random field (LMRF) model is extended to the HMRF model, and the approximate graph-cuts algorithm based on this HMRF model is derived. Experimental results show that the proposed method is more effective than the traditional graph-cuts algorithm for video moving object detection.

Moving objects detection based on kernel independent component analysis

In video moving objects detection, the same illumination and perspective, will lead to that moving objects and background is nonlinearly mixed. In this paper, Kernel Independent Component Analysis (KICA) algorithm is proposed to detect the video moving objects. By the selection of kernel's parameters, the proposed algorithm avoids the unreasonable assumption which moving objects image and background image is completely independent in the video moving objects detection based on the Independent Component Analysis (ICA). And it achieves the separation of moving objects and background in the feature space by kernel function. Experimental results demonstrate that KICA is better than ICA in video motion detection.