The Moving Target Recognition and Tracking Using RGB-D Data with the Mobile Robot

A synthetic aperture radar (SAR) automatic target recognition can effectively improve the utilization efficiency of SAR image data. In order to improve the generalization and accuracy of SAR image target recognition, a convolutional neural network model is proposed to be applied to SAR image target recognition. Firstly, the network structure is designed for the characteristics of SAR image. Secondly, the introduction of dense block network structure improves the generalization performance of the model. Finally, Dropout reduced the computational complexity of the network and improved the generalization performance of the model. Experimental data were obtained from the United States Moving and Stationary Target Acquisition and recognition database. Experimental results of 10 types of target recognition showed that the overall recognition rate of the improved convolutional neural network is 99.18%. The convolutional neural network model proposed in this study improves the accuracy and generalization of the network, and is an effective method for target recognition of SAR images.

DARPA/Air Force Research Laboratory Moving and Stationary Target Acquisition and Recognition (MSTAR) program is developing state-of-the-art model based vision approach to Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR). The model-based approach requires using off-line developed target models in an on-line hypothesize-and-test manner to compare predicted target signatures with image data and output target reports. Central to this model-based ATR is the PEMS (Predict-Extract-Match-Search) subsystem. The Search module is critical to PEMS by providing intelligent control to traverse the hypothesis feature space. A major MSTAR goal is to demonstrate robust ATR for variations in targets including partially hidden targets. This paper will provide an update on the technology being developed under MSTAR and the status of this model based ATR research, specifically concentrating on the Search Module.

A synthetic aperture radar (SAR) target recognition method was proposed based on multivariate empirical mode decomposition (MEMD). MEMD was the general extension of traditional EMD, which could avoid the mode mixing problems. MEMD was employed to process SAR images to obtain the multi-layer intrinsic mode functions (IMF), which could better reflect the time-frequency properties of the targets. Different layers of IMFs could effectively complement each other while sharing some inner correlations because they are generated from the same target. In the classification phase, the joint sparse representation was employed to represent the IMFs. The joint sparse representation could solve several related sparse representation tasks based on the idea of multi-task learning. It could produce more precise estimations than the solutions of single tasks. According to the sparse coefficient vectors corresponding to different IMFs, the reconstruction errors of different classes for the representation of the test sample can be calculated. Afterwards, the target label of the test sample can be determined. Experiments were conducted on the Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset, by comparison with existing methods under the standard operating condition, depression angle variance, noise corruption, and target occlusion, the results confirm the validity of the proposed method.

In this study, three decision fusion strategies in synthetic aperture radar (SAR) image target recognition for classification of ground vehicles in the Moving and Stationary Target Acquisition and Recognition public release database are proposed. These strategies include multi-view decision fusion strategy, multi-feature decision fusion strategy and multi-classifier decision fusion strategy. In multi-view decision fusion strategy, each view implements its recognition process individually and provides only its decision from the classifier to the fuser. In multi-feature decision fusion strategy, a variety of feature extraction algorithms are used to extract features from an image separately and each of the extracted feature vectors is submitted to a classifier. The decisions from the classifiers are then provided to the fuser. In multi-classifier decision fusion strategy, different classifiers are used to classify a feature vector. Each classifier executes its classification process individually and then provides its decision to the fuser. The fuser combines decisions to produce a single fused final decision in these three strategies. The three proposed strategies are evaluated and verified with experiments. Experimental results indicate that there are significant target recognition performance benefits in the probability of correct classification, when the proposed decision fusion strategies are properly applied in SAR image target recognition.

The underwater sensor network is currently a hot research field in academia and industry with many underwater applications, such as ocean monitoring, seismic monitoring, environment monitoring, and seabed exploration. Underwater target tracking is a critical component of ocean development. This paper studies the underwater target tracking problem of the wireless sensor network. The core technology of the target tracking algorithm is the filtering algorithm, which identifies the accuracy of the target tracking system. Nonlinear filtering is a hot issue in target tracking because feasible projects are mostly non-linear systems. The linearization method used in traditional Kalman filtering has serious shortcomings. Therefore, this paper presents the improved cubature Kalman filtering (ICKF) algorithm for underwater target tracking. There is uncertainty in the target movement, an adaptive forgetting factor is given into the cubature Kalman filtering algorithm to directly modify the error covariance to reduce the impact of uncertainties. Then, interactive multi-model technology is introduced to establish the IMMICKF algorithm with multiple states. Compared with other filtering algorithms, the new algorithm can effectively deal with non-linear target tracking problems and obtain better estimation accuracy. The numerical simulation is given to demonstrate the effectiveness of the IMMICKF algorithm.

This paper aims to increase the Unmanned Aerial Vehicle's (UAV) capacity for target tracking. First, a control model based on fuzzy logic is created, which modifies the UAV's flight attitude in response to the target's motion status and changes in the surrounding environment. Then, an edge computing-based target tracking framework is created. By deploying edge devices around the UAV, the calculation of target recognition and position prediction is transferred from the central processing unit to the edge nodes. Finally, the latest Vision Transformer model is adopted for target recognition, the image is divided into uniform blocks, and then the attention mechanism is used to capture the relationship between different blocks to realize real-time image analysis. To anticipate the position, the particle filter algorithm is used with historical data and sensor inputs to produce a high-precision estimate of the target position. The experimental results in different scenes show that the average target capture time of the algorithm based on fuzzy logic control is shortened by 20% compared with the traditional proportional-integral-derivative (PID) method, from 5.2 s of the traditional PID to 4.2 s. The average tracking error is reduced by 15%, from 0.8 m of traditional PID to 0.68 m. Meanwhile, in the case of environmental change and target motion change, this algorithm shows better robustness, and the fluctuation range of tracking error is only half of that of traditional PID. This shows that the fuzzy logic control theory is successfully applied to the UAV target tracking field, which proves the effectiveness of this method in improving the target tracking performance.

Recently, various autonomous mobile robots have been developed for practical use. To support the coexistence of robots and humans in real environments, we propose a concept named ‘Region with Velocity Constraints (RVC),’ which is set around hazardous areas. RVCs are regions where the velocities of the robot are constrained to predefined values. Inside the RVCs, the robot has to reduce its translational velocity to avoid predicted hazards such as collisions with obstacles, and to reduce its rotational velocity to prevent undesirable motions such as sharp turns. We also propose a motion planning method for navigating the mobile robot in an environment with RVCs based on the Navigation Function and Global Dynamic Window Approach. Our method generates a trajectory satisfying both translational and rotational velocity constraints to be compatible with the surroundings. Moreover, to demonstrate the validity of our method, we performed numerical simulations and experiments.

This paper proposes an automatic target recognition (ATR) method for synthetic aperture radar (SAR) images based on information-decoupled representation. A typical SAR image of a ground target can be divided into three parts: target region, shadow and background. From the aspect of SAR target recognition, the target region and shadow contain discriminative information. However, they also include some confusing information because of the similarities of different targets. The background mainly contains redundant information, which has little contribution to the target recognition. Because the target segmentation may impair the discriminative information in the target region, the relatively simpler shadow segmentation is performed to separate the shadow region for information decoupling. Then, the information-decoupled representations are generated, i.e., the target image, shadow and original image. The background is retained in the target image, which represents the coupling of target backscattering and background. The original image and generated target image are classified using the sparse representation-based classification (SRC). Then, their classification results are combined by a score-level fusion for target recognition. The shadow image is not used because of its lower discriminability and possible segmentation errors. To evaluate the performance of the proposed method, extensive experiments are conducted on the Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset under both standard operating condition (SOC) and various extended operating conditions (EOCs). The proposed method can correctly classify 10 classes of targets with the percentage of correct classification (PCC) of 94.88% under SOC. With the PCCs of 93.15% and 75.03% under configuration variance and 45° depression angle, respectively, the superiority of the proposed is demonstrated in comparison with other methods. The robustness of the proposed method to both uniform and nonuniform shadow segmentation errors is validated with the PCCs over 93%. Moreover, with the maximum average precision of 0.9580, the proposed method is more effective than the reference methods on outlier rejection.

Indoor target tracking is a crucial application of indoor localization systems. Wi-Fi fingerprint-based target tracking/positioning has been extensively studied due to its deployability under pervasive indoor wireless local area networks. However, the majority of fingerprint-based schemes adopted the static radio map and did not make full use of continuous motion information of the target. Therefore, we build a ubiquitous indoor tracking/localization system that can maintain a dynamic fingerprint database at low cost in this paper. First, we propose an optimal target tracking algorithm exploiting the historical data of the target location information to help improve the accuracy, which adopts a modified particle filter algorithm to reduce the number of samples and computing overhead. Second, we present a database self-update method according to trajectory continuity, which is employed to update fingerprint database dynamically for keeping the system robust. Finally, we tested the optimal target tracking based on dynamic fingerprint algorithm (OTTDF) in a complex laboratory area with diverse target motion conditions and various obstacles, our experimental results indicate that the OTTDF scheme successfully handles complex indoor structure, including different target motion state, signal fingerprint changes caused by obstacles, simultaneously provides better performance in localization cost, and localization/tracking accuracy in indoor wireless network.

Moving target detection and recognition methods are the foundation and key of modern intelligent video recognition systems. It combines advanced technologies in many fields such as image processing, pattern recognition, and artificial intelligence, and is a research hotspot in computer vision technology. Therefore, it is of great significance to study moving target detection and recognition algorithms. The non-local image extraction algorithm proposed in this paper uses an adaptive clustering method to perform fine cluster analysis on non-local image blocks with different feature types. Through the step-by-step principal component approximation method, we carefully find the features in each class. This progressive principal component approximation implements singular value hard threshold processing based on the Marchenko-Pastur (MP) theorem to select the main part of the feature, and uses special soft thresholds in the principal component transform domain to further improve the extraction performance. The Lower Bound-Based Within-Class Maximum Division (LBWCMD) is proposed, and this method is used as a preprocessing step of robust principal component analysis in moving target detection. This article applies LBWCMD to the video frame set based on the position information of the moving target, and the obtained frame subset meets the signal requirements of Robust Principle Component Analysis (RPCA) to the greatest extent. On this basis, we add frames with smaller motion amplitudes to each frame subset to increase the proportion of background pixels in each subset. Frame set division and low-rank decomposition realize the detection of moving targets under a unified framework. The detection rate of the proposed method is higher than that of the current popular methods in sports video data sets, and the detection accuracy is improved compared with the original RPCA method.

Feature expression is a crucial part of the target tracking process. The artificial feature is relatively simple and has strong real-time performance, but there is a problem of insufficient representation ability. It is prone to drift when dealing with problems such as rapid change and target occlusion. With the strong feature expression ability of deep neural network features in target detection and recognition tasks, deep neural network features are gradually used as feature extraction tools, but how to use and integrate these features is still worth studying. In this paper, the Residual Neural Network(ResNet) is the main researched object, and the influence of each layer on the target tracking performance is analyzed in detail. The feature fusion strategy of the convolutional layer and the addition layer is finally determined. We train a classifier separately for these layers. Then we search the multi-layer response maps to infer the target location in a coarse-to-fine fashion. The algorithm of this paper is verified on the OTB-50 dataset. The one-pass evalution(OPE) value can reach 0.612, which is better than the same type of algorithms.

Due to the extensive information in multi-view images, multi-view synthetic aperture radar (SAR) automatic target recognition (ATR) has attracted much attention. However, most current algorithms ignore the integration of multi-view topological features intrinsically related to the characteristics of SAR images. Moreover, as well as not thoroughly exploring the inherent coupling relationship, the multi-view feature fusion modules in these algorithms are also prone to cause the serve parameter burden and insufficient generality of the entire ATR model. To tackle these issues, an ATR model called multi-view visual and topological feature coordination aggregation (MVT-CA) is proposed. First, two parallel feature extraction modules are employed to extract multi-view visual and topological features independently. Specifically, the topological feature extraction module (TFM) based on a hypergraph neural network is designed to extract the implicit topological features by aggregating the context refinement features of key points within the SAR images. Subsequently, a parameterfriendly feature coordination aggregation module (FCAM) with visual and topological consistency is introduced, which effectively integrates multi-view features to generate a unified representation for classification while enhancing the generality of the entire ATR model. Experimental results on the Moving and Stationary Target Recognition (MSTAR) and the Full Aspect Stationary TargetsVehicle (FAST-Vehicle) datasets verify the effectiveness of our MVTCA model, even in scenarios involving severe background noise and target deformation

Automatic target recognition (ATR) is a mature but active research area. In an earlier paper, we proposed a novel ATR approach for recognition of targets varying in fine details, rotation, and translation using a Learning Vector Quantization (LVQ) Neural Network (NN). The proposed approach performed segmentation of multiple objects and the identification of the objects using LVQNN. In this current paper, we extend the previous approach for recognition of targets varying in rotation, translation, scale, and combination of all three distortions. We obtain the analytical results of the system level design to show that the approach performs well with some constraints. The first constraint determines the size of the input images and input filters. The second constraint shows the limits on amount of rotation, translation, and scale of input objects. We present the simulation verification of the constraints using DARPA's Moving and Stationary Target Recognition (MSTAR) images with different depression and pose angles. The simulation results using MSTAR images verify the analytical constraints of the system level design.

The existing tracking and recognition methods concentrate mainly on single‐class targets; however, systems for traffic management or intelligent transport often require multi‐class target tracking and recognition in real time. This study proposes an effective multi‐class moving target recognition method that is based on Gaussian mixture part‐based model, which accurately locates objects of interest and recognises their corresponding categories. The method is multi‐threaded and combines soft clustering approach with multiple mixture part based models to provide stable multi‐class target tracking and recognition in video sequences. The highlight of the method is its ability to recognise multi‐class moving targets and to count their numbers in the video sequence captured by a stationary camera with fixed focal length. Another contribution of this study is that an extended part based model is developed for object recognition in real‐world environments, which can improve the overall system performance, lower time costs, and better meet the actual demand of a video system. Experimental results show that the proposed method is viable in real‐time multi‐class moving target tracking and recognition.

In this paper, a novel genetic algorithm based approach to path planning of a mobile robot is proposed. The major characteristic of the proposed algorithm is that the chromosome has a variable length. The location target and obstacles are included to find a path for a mobile robot in an environment that is a 2D workplace discretized into a grid net. Each cell in the net is a gene. The number of genes in one chromosome depends on the environment. The locations of the robot, the target and the obstacle are marked in the workplace. The proposed algorithm is capable of generating collision-free paths for a mobile robot in both static and dynamic environments. In a static environment, the generated robot path is optimal in the sense of the shortest distance. The effectiveness of the proposed model is demonstrated by simulation studies.