Multi-Target Tracking and Occlusion Handling With Learned Variational Bayesian Clusters and a Social Force Model

An LP-based algorithm for the data association problem in multitarget tracking

Social force model is one of the well-known approaches that can successfully simulate pedestrians’ movements realistically. However, it is not suitable to simulate high-density crowd movement realistically due to the model having only three basic crowd characteristics which are goal, attraction, and repulsion. Therefore, it does not satisfy the high-density crowd condition which is complex yet unique, due to its capacity, density, and various demographic backgrounds of the agents. Thus, this research proposes a model that improves the social force model by introducing four new characteristics which are gender, walking speed, intention outlook, and grouping to make simulations more realistic. Besides, the high-density crowd introduces irregular behaviours in the crowd flow, which is stopping motion within the crowd. To handle these scenarios, another model has been proposed that controls each agent with two different states: walking and stopping. Furthermore, the stopping behaviour was categorized into a slow stop and sudden stop. Both of these proposed models were integrated to form a high-density crowd simulation framework. The framework has been validated by using the comparison method and fundamental diagram method. Based on the simulation of 45,000 agents, it shows that the proposed framework has a more accurate average walking speed (0.36 m/s) compared to the conventional social force model (0.61 m/s). Both of these results are compared to the real-world data which is 0.3267 m/s. The findings of this research will contribute to the simulation activities of pedestrians in a highly dense population.

This paper describes the use of a simulated real time system of a feed-forward neural network, a recurrent neural network and a set of expert rules in solving the problem of multiple target tracking. It is assumed that data is provided in the from of blips, taken off 3 consecutive focal plane arrays, operating at visible or infrared wavelengths. In this paper, the task of multiple target tracking is transformed from one of blip-frame data association to one of target clustering, which in turn is broken down and solved in four stages. Each stage is described and mapped with the use of a feed-forward, a recurrent neural network or a set of fuzzy rules. The first and second stages of the solution procedure involve the use of two feed-forward neural network modules, while the third and forth stages use a recurrent neural network module and a set of expert rules module. The multiple target tracking solution procedure is simulated through use of FORTRAN code. In principle the number of targets that can be tracked with the routine is unlimited. However, in reality, the number of targets is dictated by the number of neurons, which in turn is constrained by hardware requirements. Software simulation results shows that the multiple target tracking code is capable of tracking an arbitrary number of targets very efficiently. The program was tested and debugged for use in the tracking of sets of multiple targets; ranging from 2 to 14. Results indicated that once the average acceleration of the targets is adequately evaluated, track files could be developed with 100% accuracy.

This paper presents a method to track multiple moving humans using Ultra-Wideband (UWB) radar. UWB radar can complement other human tracking technologies, as it works well in poor visibility conditions. Our tracking approach is based on a point process interpretation of the multi-path UWB radar scattering model for moving humans. Based on this model, we present a multiple hypothesis tracking (MHT) framework for tracking the ranges and velocities of a variable number of moving human targets. The multi-target tracking (MTT) problem for UWB radar differs from traditional applications because of the complex multipath scattering observations per target. We develop an MHT framework for UWB radar-based multiple human target tracking, which can simultaneously solve the complex observation clustering and data association problems using Bayesian inference. We present experimental results in which a monostatic UWB radar tracks both individual and multiple human targets to estimate target ranges and velocities, even with changing numbers of targets across radar scans.

Human motion models are finding an increasing number of novel applications in many different fields, such as building design, computer graphics and robot motion planning. The Social Force Model is one of the most popular alternatives to describe the motion of pedestrians. By resorting to a physical analogy, individuals are assimilated to point-wise particles subject to social forces which drive their dynamics. Such a model implicitly assumes that humans move isotropically. On the contrary, empirical evidence shows that people do have a preferred direction of motion, walking forward most of the time. Lateral motions are observed only in specific circumstances, such as when navigating in overcrowded environments or avoiding unexpected obstacles. In this paper, the Headed Social Force Model is introduced in order to improve the realism of the trajectories generated by the classical Social Force Model. The key feature of the proposed approach is the inclusion of the pedestrians’ heading into the dynamic model used to describe the motion of each individual. The force and torque representing the model inputs are computed as suitable functions of the force terms resulting from the traditional Social Force Model. Moreover, a new force contribution is introduced in order to model the behavior of people walking together as a single group. The proposed model features high versatility, being able to reproduce both the unicycle-like trajectories typical of people moving in open spaces and the point-wise motion patterns occurring in high density scenarios. Extensive numerical simulations show an increased regularity of the resulting trajectories and confirm a general improvement of the model realism.

We study the pedestrian motion along a corridor in a nonpanic regime, as usually happens in evacuation scenarios in, e.g., schools, hospitals, or airports, by means of Monte Carlo simulations. We present a model, a combination of the well-known social force model (SFM) and Vicsek model (VM), that takes into account both model interactions, based on the relative position (SFM) and based on the velocity of the particles with some randomness (modulated by an external control parameter, the noise η, VM), respectively. To clarify the influence of the model ingredients we have compared simulations using (a) the pure Vicsek model (VM) with two boundary conditions (periodic and bouncing back) and with or without desired direction of motion, (b) the social force model (SFM), and (c) the model (SFM+VM). The study of steady-state particle configurations in the VM with confined geometry shows the expected bands perpendicular to the motion direction, while in the SFM and SFM+VM particles order in stripes of a given width w along the direction of motion. The results in the SFM+VM case show that w(t)≃t^{α} has a diffusivelike behavior at low noise η (dynamic exponent α≈1/2), while it is subdiffusive at high values of external noise (α<1/2). We observe the well-known order-disorder transition in the VM with both boundary conditions, but the application of a desired direction condition inhibits the existence of disorder as expected. Similar behavior is observed in the SFM case. For the SFM+VM case we find a susceptibility maximum which slowly increases with system size as a function of noise strength. This might be indicative of a order-disorder transition in the range of densities (ρε[1/12,1/9]) and speeds (v_{0}ε[0.5,2]) studied.

In the present paper, we investigate arch-shaped equilibrium solutions in the social force model proposed by Helbing and Molnar (Phys Rev E 51:4282, 1995) and Helbing et al. (Nature 407:487, 2000). The social force model is a system of ordinary differential equations, which describe the motion of the pedestrians under a panic situation. In the simulation of the social force model, we observe an intermittent appearance of arch-shaped structures (i.e. the “Blocking clusters” Parisi and Dorso, Physica A 354:606, 2005; Physica A 385:343, 2007; Frank and Dorso, Physica A 390:2135, 2011) around an exit which block up the flow of pedestrians. To understand such a dynamic behavior, we study arch-shaped equilibrium solutions around an exit under the simplest configuration.

Pedestrian movement modeling and simulation is an important way to formulate an emergency evacuation plan. The performance of the pedestrian movement model directly determines the accuracy of the simulation experiment results. Pedestrian movement in the real world is a very complicated process, and this process is susceptible to a variety of external factors (e.g. road conditions, visibility, etc.) and internal factors (e.g. psychological states, etc.) which making it difficult to model pedestrian movement. The social force model draws on the principles of mechanics and proposes a modeling scheme from the perspective of social force which greatly simplifies the process of pedestrian movement modeling. However, the traditional social force model considers the general situation of pedestrian movement and does not target the emergency evacuation scene, which has certain limitations. In this paper we start with the social force model and introduces the concepts of external guides and internal panic emotions to improves the model. In the scene, the influence of the attraction of the guide and the psychological state of the pedestrian on the movement process of the pedestrian are considered. So that the model is more suitable for the emergency evacuation scene with limited field of vision. It also provides a reference for emergency evacuation plans in public places. We take the subway station as the research object and uses Anylogic software to build simulation experiments. The experimental results show that the improved model we proposed in this paper is less than the traditional model in terms of evacuation time by 8.4%, and is better than the traditional model in terms of exit pedestrian density by 21.5%.

In real world multiple extended target tracking problems, the presence of merged measurements is a frequently occurring phenomenon, however, most existing tracking algorithms in the literature assume that each target generates independent measurements. When this measurement merging phenomenon occurs, it increases the computational complexity of the tracking algorithms. Recently, the conditional joint decision and estimation (CJDE) algorithm based on the generalized Bayes risk was proposed to solve problems of joint detection, tracking and classification (JDTC) of targets. In this paper, we develop a principled Bayesian solution to the important problem involving inter-dependent decision and estimation conditioned on data based on the theory of random finite sets (RFS), and a tractable implementation based on the recently proposed generalized labeled multi-Bernoulli (GLMB) filter. The performance of the proposed technique is demonstrated by simulation of a multi-target bearings-only tracking scenario, where measurements become merged due to finite resolution effects.

A novel particle filter for multiple target tracking with track-before-detect measurement models is proposed. Particle filters efficiently perform target tracking under nonlinear or non-Gaussian models. However, their application to multiple target tracking suffers from the curse of dimensionality. We introduce an efficient particle filter for multiple target tracking which deals with the curse of dimensionality better than previously developed methods. The proposed algorithm is tested and compared to other multiple target tracking particle filters.

This paper presents a new design for formation control on a group of autonomous humanoid robots with obstacle and collision avoidance ability that inspired by human walking behavior in a pedestrian crowd. This human walking behavior model is known as Social Force Model (SFM) and has been widely used to predict individual movement in a crowd in many variety of situation. We combine consensus algorithm with virtual structure approach for formation control and SFM for obstacle and collision avoidance. In this research, we induced some factors in SFM into consensus equation and then implemented into a group of humanoid robots. The aim of the integration is to create a group of robots that are capable of carrying out its collective tasks while still able to maintain its safety. The use of perception effect, distinct obstacle and collision avoidance function on the algorithm will differentiate algorithm with the others. The attractive feature of this new design is the various behaviors of robots to go back to their initial formation after they avoid obstacles. In the case of robot is trapped in a crowded situation (singularity condition), robot will still trying to look a new position, until it find a condition that allow him to move forward. We verify the proposed algorithm by some simulations and experiments.

In a multitarget tracking (MTT) scenario, the prior knowledge of parameters such as clutter rate and detection probability are usually uncertain or estimated offline from training data. However, significant parameters mismatch in the clutter and detection model will result in biased estimates. In such cases, the ability to adaptively online estimate clutter rate and detection probability is critical in MTT under the random finite sets (RFS) framework. In this article, we propose a robust Poisson multi-Bernoulli mixture (PMBM) filter that can accommodate model mismatch in clutter rate and detection probability for multiple extended target tracking (ETT) problems. Moreover, the closed-form solution to the proposed method is derived by the use of Beta and Gamma Gaussian inverse-Wishart (GGIW) distribution, where Beta is used to describe unknown detection probability and GGIW is used to model extended target extension as an ellipse. Simulation results are presented to verify the effectiveness of the proposed method.

Even though pedestrian motion may look chaotic in most of the cases, recent studies have shown that this motion is mainly ruled by environment and social aspects. In this paper, we propose an interacting multiple model pedestrian tracking framework that incorporates these semantic considerations as a prior knowledge about intentions and interactions between targets. We consider 4 cases of motion for pedestrians: going straight, finding one's way, walking around and standing still. Those models are competing within an Interacting Multiple Model Particle Filter strategy. Targets interactions are handled with social forces, included as potential functions in the weighting process of the Particle Filter (PF). We use different social force models in each motion model to handle high level behaviors (collision avoidance, flocking...). We evaluate our algorithm on challenging datasets and demonstrate that such semantic information improves the tracker performance.

Using CNN deep convolutional neural network model to extract image sample features, this paper designs an image multi-target tracking algorithm combining particle filter and Softmax classifier. The possible transformation state factors of the tracking target are considered in the design of the particle filter, the possible transformation states are defined by the Bayesian probability operation method, the particle filter is introduced into the CNN model, and the extracted particle image block is analyzed by the Softmax classifier. Features to achieve accurate classification. The single-target and multi-target tracking accuracy of the proposed algorithm is evaluated through four experiments. The experimental results verify that the premise of improving the accuracy of image multi-target tracking is to obtain accurate feature information of the tracked image and achieve accurate classification. The accuracy of single-target and multi-target tracking has reached more than 94%.

Pushing and overtaking others in a spatial game of exit congestion