3D Vehicle Research Articles

Application of Porcine Kidney-Derived Extracellular Matrix as Coating, Hydrogel, and Scaffold Material for Renal Proximal Tubular Epithelial Cell.

Background Human renal proximal tubular epithelial (RPTE) cell is a very useful tool for kidney-related experiments in vitro/ex vivo. However, only a few primary RPTE cells can be obtained through kidney biopsy, the proliferation rate of primary cell is very low, and the cultured cell properties are easily altered in artificial conditions. Thus, RPTE cell usage is very tricky; we applied porcine kidney-derived extracellular matrix (renal ECM) as coating, hydrogel, and scaffold material to increase cell proliferation and maintain cellular properties providing three-dimensional (3D) niche, which can be a valuable cell delivery vehicle. Methods Porcine renal ECM was prepared by decellularization using 1% Triton X-100, solubilized with 0.5 M acetic acid. The final protein concentration was adjusted to 10 μg/μL (pH 7.0). The efficacies as coating, hydrogel, and scaffold materials were analyzed through cell morphology, proliferation rate, renal-associated gene expressions, chemical composition, and microstructure evaluation. The efficacies as a coating material were compared with Matrigel, collagen type 1 (col1), gelatin, fibrinogen, and thrombin. After confirmation of coating effects, the effective concentration range was decided. The efficacies as hydrogel and scaffold materials were compared with hyaluronic acid (HA) and col1, respectively. Results As the coating material, renal ECM showed a higher cell proliferation rate compared to other materials, except for Matrigel. Renal-associated gene expressions were significantly enhanced in the renal ECM than other materials. Coating effect on cell proliferation was dependent on the renal ECM concentration, and the effective concentration ranged from 30 to 100 μg. As the hydrogel material, renal ECM showed a distinct inner cell network morphology and significantly increased renal-associated gene expressions, compared to HA hydrogel. As the scaffold material, renal ECM showed specific amide peaks, enhanced internal porosity, cell proliferation rate, and renal-associated gene expression compared to the col1 scaffold. Conclusions We concluded that renal ECM can be a suitable material for RPTE cell culture and usage. More practically, the coated renal ECM stimulates RPTE cell proliferation, and the hydrogel and scaffold of renal ECM provide useful 3D culture niche and cell delivery vehicles maintaining renal cell properties.

Cooperative Localization of 3D Vehicle Networks - The Dynamic Model

Cooperative localization (CL) algorithms have mainly focused on two-dimensional (2D) vehicle models and have almost uniformly employed only the kinematic model with velocities as inputs, making these algorithms unsuitable for force control problems. We have previously extended a CL algorithm to include a three-dimensional (3D) vehicle kinematic model with the use of quaternions to avoid singularities. In this paper, we extend the algorithm to a full 3D nonlinear dynamic vehicle model, again using quaternions. We evaluate the effectiveness of a CL algorithm and assess its benefits, when the full dynamic model is employed, by running a series of Monte Carlo type simulations with multiple relative pose measurements.

3D vehicle detection for unmanned driving systerm based on lidar

3D vehicle detection for unmanned driving systerm based on lidar

A precise analytic model for simulating tire's yaw marks striations

Tire’s yaw marks are characterized by oblique striations which contain important information about the braking and steering maneuvers performed by drivers in road accidents. This paper introduces an original method for realistic numerical and graphic simulation of yaw marks striations (scuffs) for the purpose of accidents reconstruction and analysis. The introduced method includes authors’ implementation of an existing 3D vehicle dynamics model using a non-rigid wheel/ground interaction model. Results are compared to commercial accident reconstruction software concerning vehicle dynamics implementation, and then examples of lateral sliding situations are discussed to illustrate the application of the proposed yaw marks striations model in road accidents reconstruction.

LP-GAN: Learning perturbations based on generative adversarial networks for point cloud adversarial attacks

With the rapid development of technologies for 3D model construction and analysis, the 3D model has been utilized in many applications, such as 3D reconstruction, object detection, and self-driving vehicles. Moreover, the development of deep learning theory has also enabled 3D model analysis to achieve revolutionary performance in many tasks. However, the adversarial robustness of these models has not received sufficient attention. Few methods focus on the generation of better adversarial samples for attacking the point cloud models that are utilized for testing the adversarial robustness. Many approaches only focus on shifting the critical points based on the perturbation metric but ignore the characteristics of adversarial points. In this work, we propose a novel generative model to produce adversarial samples. Based on the GAN, the generator can learn the representation of the adversarial points effectively, and adjust the adversarial samples to fool the point cloud models according to different point clouds. Furthermore, we hope that the adversarial samples not only fool the point cloud models but also can fool the human visual system with unnoticeable points, which is also the characteristic of the adversarial examples. Here, we propose a perception loss to improve the quality of the adversarial samples according to the original sample. Based on the generative model and perception loss, the final adversarial samples can effectively be utilized to improve the adversarial robustness of the point cloud models. Extensive attack experiments conducted on the ModelNet with state-of-the-art point cloud models demonstrate the effectiveness of our method.

3D Vehicle Trajectory Extraction Using DCNN in an Overlapping Multi-Camera Crossroad Scene.

The 3D vehicle trajectory in complex traffic conditions such as crossroads and heavy traffic is practically very useful in autonomous driving. In order to accurately extract the 3D vehicle trajectory from a perspective camera in a crossroad where the vehicle has an angular range of 360 degrees, problems such as the narrow visual angle in single-camera scene, vehicle occlusion under conditions of low camera perspective, and lack of vehicle physical information must be solved. In this paper, we propose a method for estimating the 3D bounding boxes of vehicles and extracting trajectories using a deep convolutional neural network (DCNN) in an overlapping multi-camera crossroad scene. First, traffic data were collected using overlapping multi-cameras to obtain a wide range of trajectories around the crossroad. Then, 3D bounding boxes of vehicles were estimated and tracked in each single-camera scene through DCNN models (YOLOv4, multi-branch CNN) combined with camera calibration. Using the abovementioned information, the 3D vehicle trajectory could be extracted on the ground plane of the crossroad by calculating results obtained from the overlapping multi-camera with a homography matrix. Finally, in experiments, the errors of extracted trajectories were corrected through a simple linear interpolation and regression, and the accuracy of the proposed method was verified by calculating the difference with ground-truth data. Compared with other previously reported methods, our approach is shown to be more accurate and more practical.

The Role of Smart Technologies to Improve the Utilization of Wood Uses in Engineering Applications: A Review

Nowadays, smart technology plays an important role in engineering applications to improve the quality of life. Thus, the development of natural materials and the use of nanotechnology, will give wood new properties to maximize its benefit. It is clear that there is a great challenge to prove the strength and durability of wood acquiring new features to reach innovative use that can influence the current path in many engineering applications. Therefore, this paper summarizes a review of the possibility of using nano- and smart-technologies to make the most of the natural and acquired potential for adding new features and physical properties of wood to improve its efficiency in architectural and mechanical applications. Moreover, experiments have shown that the use of certain types of wood in many applications such as the manufacture of 3D vehicle simulation models to study dynamic behaviors as well as in the manufacture of mechanical measurement systems to improve accuracy. In conclusion, new directions under development in this field are proposed to provide solutions to important issues in the future of measurement and quality control systems that need scientific treatment.--

Simulator Coupled with Distributed Co-Simulation Protocol for Automated Driving Tests

To meet the challenges in software testing for automated vehicles, such as increasing system complexity and an infinite number of operating scenarios, new simulation methods must be developed. Closed-loop simulations for automated driving (AD) require highly complex simulation models for multiple controlled vehicles with their perception systems as well as their surrounding context. For the realization of such models, different simulation domains must be coupled with co-simulation. However, widely supported model integration standards such as functional mock-up interface (FMI) lack native support for distributed platforms, which is a key feature for AD due to the computational intensity and platform exclusivity of certain models. The newer FMI companion standard distributed co-simulation protocol (DCP) introduces platform coupling but must still be used in conjunction with AD co-simulations. As part of an assessment framework for AD, this paper presents a DCP compliant implementation of an interoperable interface between a 3D environment and vehicle simulator and a co-simulation platform. A universal Python wrapper is implemented and connected to the simulator to allow its control as a DCP slave. A C-code-based interface enables the co-simulation platform to act as a DCP master and to realize cross-platform data exchange and time synchronization of the environment simulation with other integrated models. A model-in-the-loop use case is performed with the traffic simulator CARLA running on a Linux machine connected to the co-simulation master xMOD on a Windows computer via DCP. Several virtual vehicles are successfully controlled by cooperative adaptive cruise controllers executed outside of CARLA. The standard compliance of the implementation is verified by exemplary connection to prototypic DCP solutions from 3rd party vendors. This exemplary application demonstrates the benefits of DCP compliant tool coupling for AD simulation with increased tool interoperability, reuse potential, and performance.

3D Detection and Pose Estimation of Vehicle in Cooperative Vehicle Infrastructure System

Three-dimensional (3D) object detection is of great significance for avoiding collisions between vehicles and obstacles in autonomous driving. In particular, the recent 3D object detection methods based on supervised learning are widely studied to achieve excellent performance. However, the 3D labels for training in such methods are expensive and often difficult to be collected. To solve this issue, we propose a monocular 3D vehicle detection method. First, we propose a general mathematical K-means-like method for clustering arbitrary object contours into linear equations. Second, the position, orientation and dimensions of the vehicle can be estimated by applying K-means-like method without the need for 3D labels in the contour of the vehicle. Finally, given the 2D object detection, we maximize a posterior probability of vehicle position, orientation and dimensions to improve the accuracy of the 3D object detection based on the results of K-means-like method. We evaluate the proposed algorithm on the dataset collected by the vehicle-side and road-side cameras in the cooperative vehicle infrastructure system (CVIS). Compared with the state-of-art Deep3DBox and SMOKE methods, the evaluated results show that the detection accuracy of 3D object of our method is 1.4% higher than that of Deep3DBox in the vehicle-side system, while for the road-side camera, the proposed method has 3.86% and 4.37% higher accuracy than Deep3DBox and SMOKE, respectively. Thus, the proposed method can be seen as an effective 3D object detection method in the intelligent transportation system and CVIS.

Sight distance analyses for autonomous vehicles in Civil 3D

Abstract Nowadays, self-driving cars have a wide reputation among people that is constantly increasing, many manufacturers are developing their own autonomous vehicles. These vehicles are equipped with various sensors that are placed at several points in the car. These sensors provide information to control the vehicle (partially or completely, depending on the automation level). Sight distances on roads are defined according to various traffic situations (stopping, overtaking, crossing, etc.). Safety reasons require these sight distances, which are calculated from human factors (e.g., reaction time), vehicle characteristics (e.g., eye position, brakes), road surface properties, and other factors. Autodesk Civil 3D is a widely used tool in the field of road design, the software however was developed based on the characteristics of the human drivers and conventional vehicles.

Path Planning in Dynamic Environments with Adaptive Dimensionality

Path planning in the presence of dynamic obstacles is a challenging problem due to the added time dimension in search space. In approaches that ignore the time dimension and treat dynamic obstacles as static, frequent re-planning is unavoidable as the obstacles move, and their solutions are generally sub-optimal and can be incomplete. To achieve both optimality and completeness, it is necessary to consider the time dimension during planning. The notion of adaptive dimensionality has been successfully used in high-dimensional motion planning such as manipulation of robot arms, but has not been used in the context of path planning in dynamic environments. In this paper, we apply the idea of adaptive dimensionality to speed up path planning in dynamic environments for a robot with no assumptions on its dynamic model. Specifically, our approach considers the time dimension only in those regions of the environment where a potential collision may occur, and plans in a low-dimensional state-space elsewhere. We show that our approach is complete and is guaranteed to find a solution, if one exists, within a cost sub-optimality bound. We experimentally validate our method on the problem of 3D vehicle navigation (x, y, heading) in dynamic environments. Our results show that the presented approach achieves substantial speedups in planning time over 4D heuristic-based A*, especially when the resulting plan deviates significantly from the one suggested by the heuristic.

Path Planning with Adaptive Dimensionality

Path planning quickly becomes computationally hard as the dimensionality of the state-space increases. In this paper, we present a planning algorithm intended to speed up path planning for high-dimensional state-spaces such as robotic arms. The idea behind this work is that while planning in a high-dimensional state-space is often necessary to ensure the feasibilityof the resulting path, large portions of the path have a lower-dimensional structure. Based on this observation, our algorithm iteratively constructs a state-space of an adaptive dimensionality--a state-space that is high-dimensional only where the higher dimensionality is absolutely necessary for finding a feasible path. This often reduces drastically the size of the state-space, and as a result, the planning time and memory requirements. Analytically, we show that our method is complete and is guaranteed to find a solution if one exists, within a specified suboptimality bound. Experimentally, we apply the approach to 3D vehicle navigation (x, y, heading), and to a 7 DOF robotic arm on the Willow Garage’s PR2 robot. The results from our experiments suggest that ourmethod can be substantially faster than some of the state-of-the-art planning algorithms optimized for those tasks.

Advanced loading constraints for 3D vehicle routing problems

Given automated order systems, detailed characteristics of items and vehicles enable the detailed planning of deliveries including more efficient and safer loading of distribution vehicles. Many vehicle routing approaches ignore complex loading constraints. This paper focuses on the comprehensive evaluation of loading constraints in the context of combined Capacitated Vehicle Routing Problem and 3D Loading (3L-CVRP) and its extension with time windows (3L-VRPTW). To the best of our knowledge, this paper considers the currently largest number of loading constraints meeting real-world requirements and reducing unnecessary loading efforts for both problem variants. We introduce an approach for the load bearing strength of items ensuring a realistic load distribution between items. Moreover, we provide a new variant for the robust stability constraint enabling better performance and higher stability. In addition, we consider axle weights of vehicles to prevent overloaded axles for the first time for the 3L-VRPTW. Additionally, the reachability of items, balanced loading and manual unloading of items are taken into account. All loading constraints are implemented in a deepest-bottom-left-fill algorithm, which is embedded in an outer adaptive large neighbourhood search tackling the Vehicle Routing Problem. A new set of 600 instances is created, published and used to evaluate all loading constraints in terms of solution quality and performance. The efficiency of the hybrid algorithm is evaluated by three well-known instance sets. We outperform the benchmarks for most instance sets from the literature. Detailed results and the implementation of loading constraints are published online.

RCBi-CenterNet: An Absolute Pose Policy for 3D Object Detection in Autonomous Driving

3D Object detection is a critical mission of the perception system of a self-driving vehicle. Existing bounding box-based methods are hard to train due to the need to remove duplicated detections in the post-processing stage. In this paper, we propose a center point-based deep neural network (DNN) architecture named RCBi-CenterNet that predicts the absolute pose for each detected object in the 3D world space. RCBi-CenterNet is composed of a recursive composite network with a dual-backbone feature extractor and a bi-directional feature pyramid network (BiFPN) for cross-scale feature fusion. In the detection head, we predict a confidence heatmap that is used to determine the position of detected objects. The other pose information, including depth and orientation, is regressed. We conducted extensive experiments on the Peking University/Baidu-Autonomous Driving dataset, which contains more than 60,000 labeled 3D vehicle instances from 5277 real-world images, and each vehicle object is annotated with the absolute pose described by the six degrees of freedom (6DOF). We validated the design choices of various data augmentation methods and the backbone options. Through an ablation study and an overall comparison with the state-of-the-art (SOTA), namely CenterNet, we showed that the proposed RCBi-CenterNet presents performance gains of 2.16%, 2.76%, and 5.24% in Top 1, Top 3, and Top 10 mean average precision (mAP). The model and the result could serve as a credible benchmark for future research in center point-based object detection.

Spatial Positioning Method of Vehicle in Cross-Camera Traffic Scene

<p indent=0mm>In order to solve the problem of cross-camera and full-scene vehicle spatial positioning in traffic applications, a method combining full-scene perspective stitching and 3D vehicle detection is proposed. Firstly, a method of automatic camera calibration in traffic scenes is proposed. By building a camera calibration space model, the calibration parameters can be obtained and optimized automatically. Then, feature points in the overlapping area are used to transform and unify the multi-camera spatial coordinate system. In order to intuitively reflect the physical space of the whole scene, the perspective of each scene is transformed into a unified pixel-physical coordinate system. Finally, for the problem of 3D positioning of vehicles under monocular vision, an optimization model is established to construct a refined 3D bounding box by using the geometric constraints of vehicle projection. The 3D centroid is used as the vehicle spatial positioning point, and the mapping to the pixel-physical coordinate system can reflect the dynamic information of vehicles. In the experimental environment with artificial feature points and the practical application environment without artificial feature points, the spatial positioning experiment is carried out for multi-type vehicles in cross-camera scene. Experimental results show that the algorithm can solve the problem of vehicle positioning in the large cross-camera scene of traffic video surveillance. The comprehensive positioning accuracy in the selected experimental scene can reach the centimeter level with a real-time performance.

Computer-aided intelligent design using deep multi-objective cooperative optimization algorithm

Computer-aided product design means using artificial intelligent systems to automatically design multiple industrial products. This technique has been pervasively applied in multiple domains, such as 3D printing and vehicle manufacture. One challenge of computer-aided design is to incorporate deep neural network to optimally fuse multiple decisions. Multi-objective decision encapsulates many decision-making objectives and leverages deep CNNs to evaluate/optimize the fused multiple decisions. Due to the objectives of economic and social benefit, it is necessary to use a variety of criteria to deeply evaluate and optimize schemes. In this paper, we propose a novel quality-guided deep neural network and weighting scheme to achieve multi-objective decision. We leverage RBF neural network to construct objective weight assignment model. Then, a deep CNN is designed to implement the weighting task, each of which corresponds to a single decision. Our deep CNN has five layers and contains multilayer perceptrons, which indicate the fully connected networks. Each neuron in one layer is connected to all neurons in the next layer. The target of our deep weight-based model is that the multi-objective optimization can be formulated as a single-objective optimization by assigning different weights to each objective. Finally, the non-inferior solution of the multi-objective optimization is generated by updating the weights of the deep CNN during fine tuning. In our experiment, we have demonstrated that our method has the potential to facilitate a variety of applications, such as 3D reconstruction and system optimization. We believe that our proposed algorithm can guide the optimization of various intelligent system pipeline.

Blastability and Ore Grade Assessment from Drill Monitoring for Open Pit Applications

Blasting performance is influenced by mechanical and structural properties of the rock, on one side, and blast design parameters on the other. This paper describes a new methodology to assess rock mass quality from drill-monitoring data to guide blasting in open pit operations. Principal component analysis has been used to combine measurement while drilling (MWD) information from two drill rigs; corrections of the MWD parameters to minimize external influences other than the rock mass have been applied. First, a Structural factor has been developed to classify the rock condition in three classes (massive, fractured and heavily fractured). From it, a structural block model has been developed to simplify the recognition of rock classes. Video recording of the inner wall of 256 blastholes has been used to calibrate the results obtained. Secondly, a combined strength-grade factor has been obtained based on the analysis of the rock type description and strength properties from geology reports, assaying of drilling chips (ore/waste identification) and 3D unmanned aerial vehicle reconstructions of the post-blast bench face. Data from 302 blastholes, comprised of 26 blasts, have been used for this analysis. From the results, four categories have been identified: soft-waste, hard-waste, transition zone and hard-ore. The model determines zones of soft and hard waste rock (schisted sandstone and limestone, respectively), and hard ore zones (siderite rock type). Finally, the structural block model has been combined with the strength-grade factor in an overall rock factor. This factor, exclusively obtained from drill monitoring data, can provide an automatic assessment of rock structure, strength, and waste/ore identification.

One-Stage Anchor-Free 3D Vehicle Detection from LiDAR Sensors.

Recent one-stage 3D detection methods generate anchor boxes with various sizes and orientations in the ground plane, then determine whether these anchor boxes contain any region of interest and adjust the edges of them for accurate object bounding boxes. The anchor-based algorithm calculates the classification and regression label for each anchor box during the training process, which is inefficient and complicated. We propose a one-stage, anchor-free 3D vehicle detection algorithm based on LiDAR point clouds. The object position is encoded as a set of keypoints in the bird’s-eye view (BEV) of point clouds. We apply the voxel/pillar feature extractor and convolutional blocks to map an unstructured point cloud to a single-channel 2D heatmap. The vehicle’s Z-axis position, dimension, and orientation angle are regressed as additional attributes of the keypoints. Our method combines SmoothL1 loss and IoU (Intersection over Union) loss, and we apply as angle regression labels, which achieve high average orientation similarity (AOS) without any direction classification tricks. During the target assignment and bounding box decoding process, our framework completely avoids any calculations related to anchor boxes. Our framework is end-to-end training and stands at the same performance level as the other one-stage anchor-based detectors.

3D Detection for Occluded Vehicles From Point Clouds

In autonomous driving, 3D vehicle detection plays an important role in traffic safety, and the detection of occluded vehicles is equally important. Different from 2D images, the point clouds are sparse and disordered, causing an uncompetitive expression for occluded vehicles. As a result, current detection methods cannot achieve a higher performance for the detection of occluded vehicles. To alleviate this problem, we present a novel 3D detector for occluded vehicle (3DOV) detection using lidar data. The proposed method has the following contributions: different data representations, including a bird’s-eye view and voxel grids, which are fused to strongly represent the features of occluded vehicles; a novel voxel-wise convolutional encoder that leverages the hierarchical property of convolutional networks to learn features with spatially local correlation from unordered points; a multiscale detection header that removes the influence of occlusions and optimizes the backpropagation of gradients; and a slimmer 3DOV network is presented to handle the speed-accuracy tradeoff. The experiment results on the KITTI 3D vehicle-detection benchmark show that the 3DOV has achieved excellent performance in terms of accuracy and efficiency, especially for occluded vehicles. Ablation studies are provided to demonstrate the effectiveness of each designed module.

Validation of a PC-Crash Multibody Sport Bike Motorcycle Model

&lt;div class="section abstract"&gt;&lt;div class="htmlview paragraph"&gt;PC-Crash is an accident reconstruction program allowing the user to perform simulations with multibody objects that collide or interact with 3D vehicle mesh models. The multibody systems can be a pedestrian, a motorcycle, or a motorcycle with a rider. The multibody systems are comprised of individual rigid bodies connected by joints. The bodies can be of various size and stiffness along with varying coefficients of friction and restitution. Additionally, the joints can be tailored to define pivot types and range of motion. The current motorcycle models in PC-Crash are generic and do not resemble a sport bike type motorcycle. They are only globally scalable such that you cannot adjust length, width, or height independently. However, the user can adjust each body and/or joint individually as needed.&lt;/div&gt;&lt;div class="htmlview paragraph"&gt;A model was created that resembled a modern sport bike motorcycle. In addition, a multibody rider was mounted on the motorcycle in a typical sport bike riding position. The model was validated using the three instrumented tests that were the subject of previous research (SAE 2020-01-0885). The aforementioned tests involved moving motorcycles striking moving cars. Impact configurations involved the classic T-bone type accident, while others were more complex in nature involving significant longitudinal and lateral impact forces to the motorcycle.&lt;/div&gt;&lt;div class="htmlview paragraph"&gt;The test results were compared to parameters calculated in PC-Crash. Results such as Delta-V, yaw rate and overall post impact trajectories of the motorcycle, rider and movement of the target vehicle were compared to the data from the instrumented test vehicles. In total the multibody rider/motorcycle model was able to simulate the three staged crash tests very well. The overall trajectories of both the rider and motorcycle aligned very well with the overall motion seen in the crash test.&lt;/div&gt;&lt;/div&gt;