Scale recovery in multicamera cluster SLAM with non-overlapping fields of view
基于非重叠视域的多摄像机群SLAM尺度回复
Computer Vision and Image Understanding, Volume 126, September 2014, Pages 53-66
Abstract: A relative pose and target model estimation framework using calibrated multicamera clusters is presented. It is able to accurately track up-to-date relative motion, including scale, between the camera cluster and the (free-moving) completely unknown target object or environment using only image measurements from a set of perspective cameras. The cameras within the cluster may be arranged in any configuration, even such that there is no spatial overlap in their fields-of-view. An analysis of the set of degenerate motions for a cluster composed of three cameras is performed. It is shown that including the third camera eliminates many of the previously known ambiguities for two-camera clusters. The estimator performance and the degeneracy analysis conclusions are confirmed in experiment with ground truth data collected from an optical motion capture system for the proposed three-camera cluster against other camera configurations suggested in the literature.
Temporal synchronization of non-overlapping videos using known object motion
通过已知目标运动的非重叠录像时间同步化研究
Pattern Recognition Letters, Volume 32, Issue 1, 1 January 2011, Pages 38-46
Abstract: This paper presents a robust technique for temporally aligning multiple video sequences that have no spatial overlap between their fields of view. It is assumed that (i) a moving target with known trajectory is viewed by all cameras at non-overlapping periods in time, (ii) the target trajectory is estimated with a limited error at a constant sampling rate, and (iii) the sequences are recorded by stationary cameras with constant frame rates and fixed intrinsic and extrinsic parameters. The proposed approach reduces the problem of synchronizing N non-overlapping sequences to the problem of robustly estimating a single line from a set of appropriately-generated points in . This line describes all temporal relations between the N sequences and the moving target. Our technique can handle arbitrarily-large misalignments between the sequences and does not require any a priori information about their temporal relations. Experimental results with real-world and synthetic sequences demonstrate that our method can accurately align the videos.
Robust single object tracker based on kernelled patch of a fixed RGB camera
基于固定RGB摄像机内核补丁的鲁棒单一目标跟踪器
Optik - International Journal for Light and Electron Optics, Volume 127, Issue 3, February 2016, Pages 1100-1110
Abstract: One of the weaknesses of the mean-shift tracker is its limited ability to track a fast-moving object. Not only does the captured image look blurred but the object will often be out of the search window. Moreover, blur issue is a normal occurrence for a fast-moving object by using a commercially available camera due to a low frame rate. This paper aims to develop a robust observation detector for a single object tracker, especially for blurred image, fast-moving and non-rigid object. We propose a fusion of a kernel-based histogram and feature detector based approach. The measurement input is selected from candidate patches that will be generated by matching the vector descriptor. These vectors are built based on the detected points of interest. Two colour spaces are considered where all three channels of RGB are used, while only the hue channel of the HSV space is utilized. The kernel method is employed for better histogram accumulation. The output patch of the previous frame will be the target model where histogram similarity is measured based on Gaussian distribution. The selected output of the patch matching will undergo position smoothing for better precision. Maximum likelihood approach is used to iterate the patch position until the best match is found. The results indicate that our algorithm has a good detection for challenging surroundings and environments. In some cases, our algorithm may have less accuracy but in no case did it failed to detect.
Improving classification rate constrained to imbalanced data between overlapped and non-overlapped regions by hybrid algorithms
应用混合算法提高受制于重叠与非重叠区域之间不平衡数据的分类率
Neurocomputing, Volume 152, 25 March 2015, Pages 429-443
Abstract: A new aspect of imbalanced data classification was studied. Unlike the classical imbalanced data classification where the cause of problem is due to the difference of data sizes, our study concerns only the situation when there exists an overlap between two classes. When one class overlaps another class, there are three regions induced from the overlap. The first region is the overlapped region between two classes. The rest is the non-overlapped region of each class. The imbalance situation is obviously caused by the different amount of data at the overlapped region and non-overlapped region. In this situation, the difference of data sizes from different classes is not the main concern and has no effect on the accuracy of classification. In this research, a combined technique, called Soft-Hybrid algorithm, was proposed for improving classification performance. The technique was divided into two main phases: boundary region determination and responsive classification algorithms for each sub-area. In the first phase, data were grouped as (1) non-overlapping data, (2) borderline data, and (3) overlapping data. Learning data using modified Hausdorff Distance, Radial Basis Function Network and K-Means clustering technique with Mahalanobis Distance. Then, modified Kernel Learning Method, modified DBSCAN and RBF network were applied to classify the data into proper groups based on statistical values from the classification phase. Finally, the results of all techniques were combined. The experimental results illustrated that the proposed method can significantly improve the effectiveness in classifying imbalanced data having large overlapping sections based on TP rate, F-measure and G-mean measures. Moreover, the computational times of the proposed method were lower than the standard algorithms used for this type of this problem.
Object tracking across non-overlapping views by learning inter-camera transfer models
通过学习交互式摄像机转换模型的跨非重叠场景下的目标跟踪
Pattern Recognition, Volume 47, Issue 3, March 2014, Pages 1126-1137
Abstract: In this paper, we introduce a novel algorithm to solve the problem of object tracking across multiple non-overlapping cameras by learning inter-camera transfer models. The transfer models are divided into two parts according to different kinds of cues, i.e. spatio-temporal cues and appearance cues. To learn spatio-temporal transfer models across cameras, an unsupervised topology recovering approach based on N-neighbor accumulated cross-correlations is proposed, which estimates the topology of a non-overlapping multi-camera network. Different from previous methods, the proposed topology recovering method can deal with large amounts of data without considering the size of time window. To learn inter-camera appearance transfer models, a color transfer method is used to model the changes of color characteristics across cameras, which has an advantage of low requirements to training samples, making update efficient when illumination conditions change. The experiments are performed on different datasets. Experimental results demonstrate the effectiveness of the proposed algorithm.
Tracking multiple interacting targets in a camera network
基于摄像机网络的多交互式目标跟踪
Computer Vision and Image Understanding, Volume 134, May 2015, Pages 64-73
Abstract：In this paper we propose a framework for tracking multiple interacting targets in a wide-area camera network consisting of both overlapping and non-overlapping cameras. Our method is motivated from observations that both individuals and groups of targets interact with each other in natural scenes. We associate each raw target trajectory (i.e., a tracklet) with a group state, which indicates if the trajectory belongs to an individual or a group. Structural Support Vector Machine (SSVM) is applied to the group states to decide if merge or split events occur in the scene. Information fusion between multiple overlapping cameras is handled using a homography-based voting scheme. The problem of tracking multiple interacting targets is then converted to a network flow problem, for which the solution can be obtained by the K-shortest paths algorithm. We demonstrate the effectiveness of the proposed algorithm on the challenging VideoWeb dataset in which a large amount of multi-person interaction activities are present. Comparative analysis with state-of-the-art methods is also shown.
A visualization framework for team sports captured using multiple static cameras
多固定摄像机在团队运动可视化框架中的应用
Computer Vision and Image Understanding, Volume 118, January 2014, Pages 171-183
Abstract：We present a novel approach for robust localization of multiple people observed using a set of static cameras. We use this location information to generate a visualization of the virtual offside line in soccer games. To compute the position of the offside line, we need to localize players’ positions, and identify their team roles. We solve the problem of fusing corresponding players’ positional information by finding minimum weight K-length cycles in a complete K-partite graph. Each partite of the graph corresponds to one of the K cameras, whereas each node of a partite encodes the position and appearance of a player observed from a particular camera. To find the minimum weight cycles in this graph, we use a dynamic programming based approach that varies over a continuum from maximally to minimally greedy in terms of the number of graph-paths explored at each iteration. We present proofs for the efficiency and performance bounds of our algorithms. Finally, we demonstrate the robustness of our framework by testing it on 82,000 frames of soccer footage captured over eight different illumination conditions, play types, and team attire. Our framework runs in near-real time, and processes video from 3 full HD cameras in about 0.4 s for each set of corresponding 3 frames.
Efficient simulation for positioning and utilizing of multiple cameras
多摄像机定位与使用的高效模拟
Simulation Modelling Practice and Theory, Volume 34, May 2013, Pages 37-47
Abstract：This paper proposes the efficient camera placement method that considers spatial information and gives priorities of spaces in the viewpoint of people moving pattern. To efficiently cover realistic environments, camera performance and installation cost are included in our model. Simulation results show that the proposed placement method not only optimally determines the number of cameras but also coordinates the position and orientation of cameras in a utility-maximized way. Furthermore, this paper can provide a near optimal solution at a very low computational cost based on peak detection and Kullback–Leibler divergence concepts. The proposed method is evaluated and compared in terms of the computational cost and the coverage rate with the greedy approach.
Collaborative detection of repetitive behavior by multiple uncalibrated cameras
基于多未标定摄像机的重复行为协作检测
Information Fusion, Volume 21, January 2015, Pages 68-81
Abstract：In smart environments, the embedded sensing systems should intelligently adapt to the behavior of the users. Many interesting types of behavior are characterized by repetition of actions such as certain activities or movements. A generic methodology to detect and classify repetitions that may occur at different scales is introduced in this paper. The proposed method is called Action History Matrices (AHM). The properties of AHM for detecting repetitive movement behavior are demonstrated in analyzing four customer behavior classes in a shop environment observed by multiple uncalibrated cameras.
Two different datasets, video recordings in the shop environment and motion path simulations, are created and used in the experiments. The AHM-based system achieves an accuracy of 97% with most suitable scale and naive Bayesian classifier on the single-view simulated movement data. In addition, the performance of two fusion levels and three fusion methods are compared with AHM method on the multi-view recordings. In our results, fusion at the decision-level offers consistently better accuracy than feature-level, and the coverage-based view-selection fusion method (51%) marginally outperforms the majority method. The upper limit with the recorded data for accuracy by view-selection is found to be 75%.
Joint multi-person detection and tracking from overlapping cameras
基于重叠摄像机的联合多人检测与跟踪
Computer Vision and Image Understanding, Volume 128, November 2014, Pages 36-50
Abstract: We present a system to track the positions of multiple persons in a scene from overlapping cameras. The distinguishing aspect of our method is a novel, two-step approach that jointly estimates person position and track assignment. The proposed approach keeps solving the assignment problem tractable, while taking into account how different assignments influence feature measurement. In a hypothesis generation stage, the similarity between a person at a particular position and an active track is based on a subset of cues (appearance, motion) that are guaranteed observable in the camera views. This allows for efficient computation of the K-best joint estimates for person position and track assignment under an approximation of the likelihood function. In a subsequent hypothesis verification stage, the known person positions associated with these K-best solutions are used to define a larger set of actually visible cues, which enables a re-ranking of the found assignments using the full likelihood function.
We demonstrate that our system outperforms the state-of-the-art on four challenging multi-person datasets (indoor and outdoor), involving 3–5 overlapping cameras and up to 23 persons simultaneously. Two of these datasets are novel: we make the associated images and annotations public to facilitate benchmarking.
