Distributed Extended Object Tracking with Adaptive Networks

In this paper, an adaptive neural network architecture is proposed for efficient video object segmentation and tracking of stereoscopic sequences. The scheme includes (a) a retraining algorithm for adapting network weights to current conditions, (b) a semantically meaningful object extraction module for creating a retraining set and (c) a decision mechanism, which detects the time instances that a new network retraining is required. The retraining algorithm optimally adapts network weights by exploiting information of the current condition with a minimal deviation of the network weights. Description of the current conditions is provided by a segmentation fusion scheme, which appropriately combines color and depth information.

In this chapter, an adaptive neural network architecture is proposed for efficient knowledge extraction in video sequences. The system is focused on video object segmentation and tracking in stereoscopic video sequences. The proposed scheme includes: (a) a retraining algorithm for adapting the network weights to current conditions, (b) a semantically meaningful object extraction module for creating a retraining set and (c) a decision mechanism, which detects the time instances when a new network retraining is activated. The retraining algorithm optimally adapts network weights by exploiting information of the current conditions with a minimal deviation of the network weights. The algorithm results in the minimization of a convex function subject to linear constraints, and thus, one minimum exists. Description of current conditions is provided by a segmentation fusion scheme, which appropriately combines color and depth information. Experimental results on real-life video sequences are presented to indicate the promising performance of the proposed adaptive neural network-based scheme.

Handling occlusion in object tracking in stereoscopic video sequences

PHD filtering is a common and effective multiple object tracking (MOT) algorithm used in scenarios where the number of objects and their states are unknown. In scenarios where each object can generate multiple measurements per scan, some PHD filters can estimate the extent of the objects as well as their kinematic properties. Most of these approaches are, however, not able to inherently estimate trajectories and rely on ad-hoc methods, such as different labeling schemes, to build trajectories from the state estimates. This paper presents a Gamma Gaussian inverse Wishart mixture PHD filter that can directly estimate sets of trajectories of extended targets by expanding previous research on tracking sets of trajectories for point source objects to handle extended objects. The new filter is compared to an existing extended PHD filter that uses a labeling scheme to build trajectories, and it is shown that the new filter can estimate object trajectories more reliably.

NN-adaptive predictive control for a class of discrete-time nonlinear systems with input-delay

A literature analysis has shown that object search, recognition, and tracking systems are becoming increasingly popular. However, such systems do not achieve high practical results in analyzing small moving living objects ranging from 8 to 14 mm. This article examines methods and tools for recognizing and tracking the class of small moving objects, such as ants. To fulfill those aims, a customized You Only Look Once Ants Recognition (YOLO_AR) Convolutional Neural Network (CNN) has been trained to recognize Messor Structor ants in the laboratory using the LabelImg object marker tool. The proposed model is an extension of the You Only Look Once v4 (Yolov4) 512 × 512 model with an additional Self Regularized Non–Monotonic (Mish) activation function. Additionally, the scalable solution for continuous object recognizing and tracking was implemented. This solution is based on the OpenDatacam system, with extended Object Tracking modules that allow for tracking and counting objects that have crossed the custom boundary line. During the study, the methods of the alignment algorithm for finding the trajectory of moving objects were modified. I discovered that the Hungarian algorithm showed better results in tracking small objects than the K–D dimensional tree (k-d tree) matching algorithm used in OpenDataCam. Remarkably, such an algorithm showed better results with the implemented YOLO_AR model due to the lack of False Positives (FP). Therefore, I provided a new tracker module with a Hungarian matching algorithm verified on the Multiple Object Tracking (MOT) benchmark. Furthermore, additional customization parameters for object recognition and tracking results parsing and filtering were added, like boundary angle threshold (BAT) and past frames trajectory prediction (PFTP). Experimental tests confirmed the results of the study on a mobile device. During the experiment, parameters such as the quality of recognition and tracking of moving objects, the PFTP and BAT, and the configuration parameters of the neural network and boundary line model were analyzed. The results showed an increased tracking accuracy with the proposed methods by 50%. The study results confirmed the relevance of the topic and the effectiveness of the implemented methods and tools.

Extended object tracking (EOT) is gaining momentum in recent years. The random matrix method is a popular EOT method, which has a simple yet effective framework. The existing random matrix approaches usually assume that scatter centers are uniformly or symmetrically distributed on the object. However, due to weak signal strength and limited measuring angles, scatter centers are densely distributed on several (separated) parts rather than uniformly distributed on the whole object. This results in several measurement-densely-distributed parts. To address this problem, this paper proposes a new conditional Gaussian mixture model. Each component of the model characterizes a single measurement-densely-distributed part. The object extension is modeled as an ellipse by a random matrix. As a portion of the object, each measurement-densely-distributed part is modeled by scaling and shifting that ellipse and transforming that random matrix. This modeling can effectively describe the non-uniformly distributed measurements, but also inherit the simplicity of the random matrix model. Based on the new model and utilized the random matrix framework, a variational Bayesian approach is derived, which recursively and efficiently estimates kinematic and extension states. From the simulation and real experimental results, the proposed approach has a significant performance improvement over some existing random matrix approaches.

This paper presents a data-driven measurement model for extended object tracking (EOT) with automotive radar. Specifically, the spatial distribution of automotive radar measurements is modeled as a hierarchical truncated Gaussian with structural geometry parameters (e.g., truncation bounds, their orientation, and a scaling factor) learned from the training data. The contribution is twofold. First, the learned measurement model can provide an adequate resemblance to the spatial distribution of real-world automotive radar measurements. Second, large-scale offline training datasets can be leveraged to learn the geometry-related parameters and offload the computationally demanding model parameter estimation from the state update step. The learned structural measurement model is further incorporated into the random matrix-based EOT approach with a new state update step. The effectiveness of the proposed approach is verified on the nuScenes dataset.

The trend towards highly-automated driver assistance systems strongly depends on the performance of environmental perception systems. Increasingly, greater attention is paid to sensor data modeling, because more measurement information can be gained with a good understanding of the hardware. Radar sensors are cost-efficient and thus often used for kinematic object tracking. Current techniques for the utilization of high-resolution radar sensors for object extent estimation, usually model the spatial occurrence of radar detections using statistical approximations. Although providing fast calculation speeds, the obtainable accuracy is not sufficient, especially when low-cost hardware is used. In this work, we propose a radar model which is based on approximations of every functional abstraction layer of radar sensors, from the electromagnetic wave distribution to the digital signal processing. This approach allows an inherent and generic modeling of the spatial detection probability of the environment, regardless of its exact content and avoiding any complexity issues. Consequently, the interaction between measurements of multiple vehicles can be accurately modeled. Additionally, various details like the antenna diagram and underbody reflections are included. The final radar model is evaluated and compared to measurement data.

In this paper, a novel image moment-based model for extended object shape estimation and tracking is presented. A method to represent and estimate an elliptical shape using its image moments is first developed. The model of representing the shape of an object falls under the category of random hypersurface model (RHM) for extended object tracking. The moments are estimated using an unscented Kalman filter (UKF). Measurements of 2D points sampled from the object are used to estimate the shape. A measurement model for the UKF is derived based on the 2D locations of the feature points sampled from the object. Simulation results of the proposed image moment-based model are presented to estimate the shape of static ellipse, and of the moving ellipse.

An adaptive virtual measurement STGP algorithm for self-occluded 3D extended object tracking with time-varying shapes