Collision Prediction Research Articles

A Review on Cross Weather Traffic Scene Understanding Using Transfer Learning for Intelligent Transport System

Intelligent transport systems (ITS) have revolutionized the transportation industry by integrating cutting-edge technologies to enhance road safety, reduce traffic congestion and optimize the transportation network. Scene understanding is a critical component of ITS that enables real-time decision-making by interpreting the environment's contextual information. However, achieving accurate scene understanding requires vast amounts of labeled data, which can be costly and time-consuming. It is quite challenging to Understand traffic scene captured from vehicle mounted cameras. In recent times, the combination of road scene-graph representations and graph learning techniques has demonstrated superior performance compared to cutting-edge deep learning methods across various tasks such as action classification, risk assessment, and collision prediction. It's a grueling problem due to large variations under different illumination conditions. Transfer learning is a promising approach to address this challenge. Transfer learning involves leveraging pre-trained deep learning models on large-scale datasets to develop efficient models for new tasks with limited data. In the context of ITS, transfer learning can enable accurate scene understanding with less data by reusing learned features from other domains.
 This paper presents a comprehensive overview of the application of transfer learning for scene understanding in cross domain. It highlights the benefits of transfer learning for ITS and presents various transfer learning techniques used for scene understanding. This survey paper provides systematic review on cross domain outdoor scene understanding and transfer learning approaches from different perspective, presents information on current state of art and significant methods in choosing the right transfer learning model for specific scene understanding applications.

Effectiveness of the Autonomous Braking and Evasive Steering System OPREVU-AES in Simulated Vehicle-to-Pedestrian Collisions

This paper proposes a combined system (OPREVU-AES) that integrates optimized AEB and Automatic Emergency Steering (AES) to generate evasive maneuvers, and it provides an assessment of its effectiveness when compared to a commercial AEB system. The optimized AEB system regulates the braking response through a collision prediction model. OPREVU is a research project in which INSIA-UPM and CEDINT-UPM cooperate to improve driving assistance systems and to characterize pedestrians’ behavior through virtual reality (VR) techniques. The kinematic and dynamic analysis of OPREVU-AES is conducted using CarSim© software v2020.1. The avoidance trajectories are predefined for speeds above 40 km/h, which controls the speed and lateral stability during the overtaking and lane re-entry process. In addition, the decision algorithm integrates information from the lane and the blind spot detectors. The effectiveness evaluation is based on the reconstruction of a sample of vehicle-to-pedestrian crashes (INSIA-UPM database), using PCCrash© software v. 2013, and it considers the probability of head injury severity (ISP) as an indicator. The incorporation of AEB can avoid 53.8% of accidents, with an additional 2.5–3.5% avoided by incorporating automatic steering. By increasing the lateral activation range, the total avoidance rate is increased to 61.8–69.8%. The average ISP reduction is 65%, with significant reductions achieved in most cases where avoidance is not possible.

Technical note: A collision prediction tool using Blender.

Non-coplanar radiotherapy treatment techniques on C-arm linear accelerators have the potential to reduce dose to organs-at-risk in comparison with coplanar treatment techniques. Accurately predicting possible collisions between gantry, table and patient during treatment planning is needed to ensure patient safety. We offer a freely available collision prediction tool using Blender, a free and open-source 3D computer graphics software toolset. A geometric model of a C-arm linear accelerator including a library of patient models is created inside Blender. Based on the model, collision predictions can be used both to calculate collision-free zones and to check treatment plans for collisions. The tool is validated for two setups, once with and once without a full body phantom with the same table position. For this, each gantry-table angle combination with a 2° resolution is manually checked for collision interlocks at a TrueBeam system and compared to simulated collision predictions. For the collision check of a treatment plan, the tool outputs the minimal distance between the gantry, table and patient model and a video of the movement of the gantry and table, which is demonstrated for one use case. A graphical user interface allows user-friendly input of the table and patient specification for the collision prediction tool. The validation resulted in a true positive rate of 100%, which is the rate between the number of correctly predicted collision gantry-table combinations and the number of all measured collision gantry-table combinations, and a true negative rate of 89%, which is the ratio between the number of correctly predicted collision-free combinations and the number of all measured collision-free combinations. A collision prediction tool is successfully created and able to produce maps of collision-free zones and to test treatment plans for collisions including visualisation of the gantry and table movement.

STRIDE: Street View-based Environmental Feature Detection and Pedestrian Collision Prediction.

This paper introduces a novel benchmark to study the impact and relationship of built environment elements on pedestrian collision prediction, intending to enhance environmental awareness in autonomous driving systems to prevent pedestrian injuries actively. We introduce a built environment detection task in large-scale panoramic images and a detection-based pedestrian collision frequency prediction task. We propose a baseline method that incorporates a collision prediction module into a state-of-the-art detection model to tackle both tasks simultaneously. Our experiments demonstrate a significant correlation between object detection of built environment elements and pedestrian collision frequency prediction. Our results are a stepping stone towards understanding the interdependencies between built environment conditions and pedestrian safety.

Enhancing LGMD-based model for collision prediction via binocular structure.

Lobular giant motion detector (LGMD) neurons, renowned for their distinctive response to looming stimuli, inspire the development of visual neural network models for collision prediction. However, the existing LGMD-based models could not yet incorporate the invaluable feature of depth distance and still suffer from the following two primary drawbacks. Firstly, they struggle to effectively distinguish the three fundamental motion patterns of approaching, receding, and translating, in contrast to the natural abilities of LGMD neurons. Secondly, due to their reliance on a general determination process employing an activation function and fixed threshold for output, these models exhibit dramatic fluctuations in prediction effectiveness across different scenarios. To address these issues, we propose a novel LGMD-based model with a binocular structure (Bi-LGMD). The depth distance of the moving object is extracted by calculating the binocular disparity facilitating a clear differentiation of the motion patterns, after obtaining the moving object's contour through the basic components of the LGMD network. In addition, we introduce a self-adaptive warning depth-distance, enhancing the model's robustness in various motion scenarios. The effectiveness of the proposed model is verified using computer-simulated and real-world videos. Furthermore, the experimental results demonstrate that the proposed model is robust to contrast and noise.

An agile autonomous car driving assistance using hybrid optimization-based kernel support vector convolutional network

In robotic control systems, autonomous car driving is considered a complicated task. The conventional modular techniques necessitate precise localization, planning, and mapping procedures to provide safe driving. These requirements make autonomous car driving tasks computationally ineffective and sensitive to ecological changes. In recent times, the emergence of deep learning-based approaches in autonomous car driving systems provided promising solutions for complex vision task interpretation, localization, and environmental perception. But, they create distribution mismatch issues and require huge perilous interaction information to offer safe driving with the prediction of future collisions. Therefore, to overcome this issue, a novel Modified Kernel Support vector Convolutional based Hybrid Grasshopper Harris hawk (MKSC-HGH) approach is proposed. The proposed MKSC-HGH approach efficiently forecasts the uncertainties of agile autonomous car driving and guides proper navigation details of the track to avoid the possibility of collision with on-road obstacles and fences. This system inputs camera images captured through sensors, car speed, and manual driving data to analyze and predict safety control and probable collision. The prominent feature representing the ability of Modified Convolutional Neural Network (MCNN) and complex boundary encoding process of Kernel Support Vector Machine (KSVM) with hyperparameter tuning operation of Modified Grasshopper Optimization Algorithm based Improved Harris Hawk Optimization (MGOA-IHHO) algorithm made proposed system applicable to predict future states (ie. visual prediction, speed prediction, and collision prediction) more accurately. The efficiency of the proposed MKSC-HGH approach is investigated using evaluation measures namely completion time, average speed, top speed, and laps time. The simulation outcome shows the superiority of the proposed MKSC-HGH approach in predicting agile autonomous car driving over state of art techniques about task performance and sample efficiency.

BOSS: Bones, organs and skin shape model

A virtual anatomical model of a patient can be a valuable tool for enhancing clinical tasks such as workflow automation, patient-specific X-ray dose optimization, markerless tracking, positioning, and navigation assistance in image-guided interventions. For these tasks, it is highly desirable that the patient’s surface and internal organs are of high quality for any pose and shape estimate. At present, the majority of statistical shape models (SSMs) are restricted to a small number of organs or bones or do not adequately represent the general population. To address this, we propose a deformable human shape and pose model that combines skin, internal organs, and bones, learned from CT images. By modeling the statistical variations in a pose-normalized space using probabilistic PCA while also preserving joint kinematics, our approach offers a holistic representation of the body that can be beneficial for automation in various medical applications. In an interventional setup, our model could, for example, facilitate automatic system/patient positioning, organ-specific iso-centering, automated collimation or collision prediction. We assessed our model’s performance on a registered dataset, utilizing the unified shape space, and noted an average error of 3.6 mm for bones and 8.8 mm for organs. By utilizing solely skin surface data or patient metadata like height and weight, we find that the overall combined error for bone-organ measurement is 8.68 mm and 8.11 mm, respectively. To further verify our findings, we conducted additional tests on publicly available datasets with multi-part segmentations, which confirmed the effectiveness of our model. In the diverse TotalSegmentator dataset, the errors for bones and organs are observed to be 5.10mm and 8.72mm, respectively. Our work shows that anatomically parameterized statistical shape models can be created accurately and in a computationally efficient manner. The proposed approach enables the construction of shape models that can be directly integrated into to various medical applications.

Vision Transformer Customized for Environment Detection and Collision Prediction to Assist the Visually Impaired.

This paper presents a system that utilizes vision transformers and multimodal feedback modules to facilitate navigation and collision avoidance for the visually impaired. By implementing vision transformers, the system achieves accurate object detection, enabling the real-time identification of objects in front of the user. Semantic segmentation and the algorithms developed in this work provide a means to generate a trajectory vector of all identified objects from the vision transformer and to detect objects that are likely to intersect with the user's walking path. Audio and vibrotactile feedback modules are integrated to convey collision warning through multimodal feedback. The dataset used to create the model was captured from both indoor and outdoor settings under different weather conditions at different times across multiple days, resulting in 27,867 photos consisting of 24 different classes. Classification results showed good performance (95% accuracy), supporting the efficacy and reliability of the proposed model. The design and control methods of the multimodal feedback modules for collision warning are also presented, while the experimental validation concerning their usability and efficiency stands as an upcoming endeavor. The demonstrated performance of the vision transformer and the presented algorithms in conjunction with the multimodal feedback modules show promising prospects of its feasibility and applicability for the navigation assistance of individuals with vision impairment.

The probabilistic nature of dune collisions in 2D

Abstract. Dunes are bedforms of different size and shape, appearing throughout aeolian, subaqueous and extraterrestrial environments. Collisions between dunes drive dune field evolution, and are a direct result of interacting dunes of different heights, travelling at different speeds. We perform 2D cellular automaton simulations of collisions between dune pairs migrating in a steady flow. Modelled collisions can result in either ejection, where dunes exchange mass before separating, or downstream- or upstream-dominant coalescence (merging of dunes). For each of these three elementary types of interaction, we identify the mass exchange mechanism and the distinctive intermediate morphologies. Surprisingly, we show that the collision outcome depends probabilistically on the initial dune area ratio r and can be described by a narrow sigmoidal function centred on r=1/2. Finally, we compare our simulations with laboratory experiments of dune collisions, finding good agreement concerning the intermediate morphology and the collision outcome. Our results can motivate further observational or experimental studies that validate our probabilistic collision predictions and fully determine the controls on the coalescence–ejection transition.

Review on Type of Sensors and Detection Method of Anti-Collision System of Unmanned Aerial Vehicle.

Unmanned aerial vehicle (UAV) usage is increasing drastically worldwide as UAVs are used in various industries for many applications, such as inspection, logistics, agriculture, and many more. This is because performing a task using UAV makes the job more efficient and reduces the workload needed. However, for a UAV to be operated manually or autonomously, the UAV must be equipped with proper safety features. An anti-collision system is one of the most crucial and fundamental safety features that UAVs must be equipped with. The anti-collision system allows the UAV to maintain a safe distance from any obstacles. The anti-collision technologies are of crucial relevance to assure the survival and safety of UAVs. Anti-collision of UAVs can be varied in the aspect of sensor usage and the system's working principle. This article provides a comprehensive overview of anti-collision technologies for UAVs. It also presents drone safety laws and regulations that prevent a collision at the policy level. The process of anti-collision technologies is studied from three aspects: Obstacle detection, collision prediction, and collision avoidance. A detailed overview and comparison of the methods of each element and an analysis of their advantages and disadvantages have been provided. In addition, the future trends of UAV anti-collision technologies from the viewpoint of fast obstacle detection and wireless networking are presented.

Determining Key Parameters with Data-Assisted Analysis of Conditionally Automated Driving

In conditionally automated driving, a vehicle issues a take-over request when it reaches the functional limits of self-driving, and the driver must take control. The key driving parameters affecting the quality of the take-over (TO) process have yet to be determined and are the motivation for our work. To determine these parameters, we used a dataset of 41 driving and non-driving parameters from a previous user study with 216 TOs while performing a non-driving-related task on a handheld device in a driving simulator. Eight take-over quality aspects, grouped into pre-TO predictors (attention), during-TO predictors (reaction time, solution suitability), and safety performance (off-road drive, braking, lateral acceleration, time to collision, success), were modeled using multiple linear regression, support vector machines, M5’, 1R, logistic regression, and J48. We interpreted the best-suited models by highlighting the most influential parameters that affect the overall quality of a TO. The results show that these are primarily maximal acceleration (88.6% accurate prediction of collisions) and the TOR-to-first-brake interval. Gradual braking, neither too hard nor too soft, as fast as possible seems to be the strategy that maximizes the overall TO quality. The position of the handheld device and the way it was held prior to TO did not affect TO quality. However, handling the device during TO did affect driver attention when shorter attention times were observed and drivers held their mobile phones in only one hand. In the future, automatic gradual braking maneuvers could be considered instead of immediate full TOs.

Analysis of the influence of detouring obstacle avoidance behavior on unidirectional flow

A pedestrian’s active obstacle avoidance behavior has an important impact on crowd safety in heterogeneous crowd movements, but how it affects the movement of individuals and crowd is not yet clear. This study proposes a detour anisotropic social force model (DASFM) that considers active obstacle avoidance behavior for future collision prediction. DASFM prioritizes moving obstacle pedestrians based on velocity obstacle (VO) and combines the anisotropic social force model (ASFM) to achieve the dynamic process of pedestrian active obstacle avoidance behavior. The results show that DASFM enables individuals to safely overtake while avoiding fluctuations in pedestrian speed and distance between pedestrians. The model improved abnormal “strip group” crowd aggregation, individual jostling, and congestion found in other models. It has effective performance in spatial distribution, trajectory, density evolution, and time consumption of crowd movement in unidirectional flow under open conditions. The fundamental diagram generated by DASFM in unidirectional flow under periodic boundary conditions tends to be consistent with the observations and experimental results of other researchers. Instantaneous flow analysis shows that overtaking behavior alleviates local congestion in the low-density crowd and promotes the slow flow of the high-density crowd. However, overtaking behavior also exacerbates the stop-and-go waves of the high-density crowd, posing a threat to crowd safety. This study not only effectively realizes the simulation of overtaking behavior among individuals and crowds, but also discovers the impact patterns of overtaking behavior on low- or high-density crowd. Therefore, measures should be taken to control individual overtaking behavior in high-density crowd to prevent the production of stop-and-go waves.

Risk-aware controller for autonomous vehicles using model-based collision prediction and reinforcement learning

Autonomous Vehicles (AVs) have the potential to save millions of lives and increase the efficiency of transportation services. However, the successful deployment of AVs requires tackling multiple challenges related to modeling and certifying safety. State-of-the-art decision-making methods usually rely on end-to-end learning or imitation learning approaches, which still pose significant safety risks. Hence the necessity of risk-aware AVs that can better predict and handle dangerous situations. Furthermore, current approaches tend to lack explainability due to their reliance on end-to-end Deep Learning, where significant causal relationships are not guaranteed to be learned from data.This paper introduces a novel risk-aware framework for training AV agents using a bespoke collision prediction model and Reinforcement Learning (RL). The collision prediction model is based on Gaussian Processes and vehicle dynamics, and is used to generate the RL state vector. Using an explicit risk model increases the post-hoc explainability of the AV agent, which is vital for reaching and certifying the high safety levels required for AVs and other safety-sensitive applications. Experimental results obtained with a simulator and state-of-the-art RL algorithms show that the risk-aware RL framework decreases average collision rates by 15%, makes AVs more robust to sudden harsh braking situations, and achieves better performance in both safety and speed when compared to a standard rule-based method (the Intelligent Driver Model). Moreover, the proposed collision prediction model outperforms other models in the literature.

Autonomous overtaking control considering the velocity change of obstacle vehicles on two-lane highways

Aiming at the speed change of obstacle vehicles in the process of autonomous ground vehicle (AGV) overtaking on two-lane highways, an overtaking control method including trajectory planning and trajectory control is proposed. First, the reference overtaking trajectory of the AGV is planned in stages (lane changing, passing and merging) using fifth order polynomials satisfying stability and comfort constraints. The initial planning trajectory may be unsafe due to the speed change of obstacle vehicles. So a collision prediction algorithm is built to predict two kinds of dangerous situations in real time: AGV collides with the leading vehicle during lane changing, and AGV collides with the oncoming vehicle during merging. Secondly, the potential field of the obstacle vehicles embedded with the relative speed is established and combined with the road potential field as the risk indicator. Then, combining the nonlinear model predictive control algorithm and risk index, the overtaking trajectory is replanned in real time with the initial trajectory as reference. Finally, a lateral path tracking controller based on model predictive control and a longitudinal speed tracking controller based on sliding mode control are established to track the planned trajectory rapidly and stably. To verify the performance of the proposed method, overtaking in three cases is simulated, and vehicle responding curves are obtained. The results show that the proposed control method can achieve safe and stable overtaking.

Residual gas damping noise in constrained volume in space-borne gravitational wave detection

The residual gas damping in the enclosure surrounding a free-fall test mass is an important limiting factor for the sensitivity of some space detection missions. We present a constrained volume damping model for space-borne gravitational wave detection, which depends on the escape time needed for a molecule to diffuse out of the gap between two adjacent surfaces. Using a completely inelastic molecular collision prediction, we calculate the escape times in the two most common constrained volumes, cylinder and cuboid, respectively. Both calculations agree well with Monte Carlo simulations. Additionally, the theoretical model of the residual gas damping presented in this work is in good agreement with the experimental results in the ground torsion pendulum test of LISA and its precursor LISA Pathfinder.

A robust collision prediction and detection method based on neural network for autonomous delivery robots

AbstractFor safe last‐mile autonomous robot delivery services in complex environments, rapid and accurate collision prediction and detection is vital. This study proposes a suitable neural network model that relies on multiple navigation sensors. A light detection and ranging technique is used to measure the relative distances to potential collision obstacles along the robot's path of motion, and an accelerometer is used to detect impacts. The proposed method tightly couples relative distance and acceleration time‐series data in a complementary fashion to minimize errors. A long short‐term memory, fully connected layer, and SoftMax function are integrated to train and classify the rapidly changing collision countermeasure state during robot motion. Simulation results show that the proposed method effectively performs collision prediction and detection for various obstacles.

Using Machine Learning on V2X Communications Data for VRU Collision Prediction.

Intelligent Transportation Systems (ITSs) are systems that aim to provide innovative services for road users in order to improve traffic efficiency, mobility and safety. This aspect of safety is of utmost importance for Vulnerable Road Users (VRUs), as these users are typically more exposed to dangerous situations, and their vehicles also possess poorer safety mechanisms when in comparison to regular vehicles on the road. Implementing automatic safety solutions for VRU vehicles is challenging since they have high agility and it can be difficult to anticipate their behavior. However, if equipped with communication capabilities, the generated Vehicle-to-Anything (V2X) data can be leveraged by Machine Learning (ML) mechanisms in order to implement such automatic systems. This work proposes a VRU (motorcyclist) collision prediction system, utilizing stacked unidirectional Long Short-Term Memorys (LSTMs) on top of communication data that is generated using the VEINS simulation framework (coupling the Simulation of Urban MObility (SUMO) and Network Simulator 3 (ns-3) tools). The proposed system performed well in two different scenarios: in Scenario A, it predicted 96% of the collisions, averaging 4.53 s for Average Prediction Time (s) (APT) and with a Correct Decision Percentage (CDP) of 41% and 78 False Positives (FPs); in Scenario B, it predicted 95% of the collisions, with a 4.44 s APT, while the CDP was 43% with 68 FPs. The results show the effectiveness of the approach: using ML methods on V2X data allowed the prediction of most of the simulated accidents. Nonetheless, the presence of a relatively high number of FPs does not allow for the usage of automatic safety features (e.g., emergency breaking in the passenger vehicles); thus, collision avoidance must be achieved manually by the drivers.

A VANET Collision Warning System with Cloud Aided Pliable Q-Learning and Safety Message Dissemination

Ease of self-driving technological developments revives Vehicular Adhoc Networks (VANETs) and motivates the Intelligent Transportation System (ITS) to develop novel intelligent solutions to amplify the VANET safety and efficiency. Collision warning system plays a significant role in VANET due to the avoidance of fatalities in vehicle crashes. Different kinds of collision warning systems have been designed for diverse VANET scenarios. Among them, reinforcement-based machine learning algorithms receive much attention due to the dispensable of explicit modeling about the environment. However, it is a censorious task to retrieve the Q-learning parameters from the dynamic VANET environment effectively. To handle such issue and safer the VANET driving environment, this paper proposes a cloud aided pliable Q-Learning based Collision Warning Prediction and Safety message Dissemination (QCP-SD). The proposed QCP-SD integrates two mechanisms that are pliable Q-learning based collision prediction and Safety alert Message Dissemination. Firstly, the designing of pliable Q-learning parameters based on dynamic VANET factors with Q-learning enhances collision prediction accuracy. Further, it estimates the novel metric named as Collision Risk Factor (CRF) and minimizes the driving risks due to vehicle crashes. The execution of pliable Q-learning only at RSUs minimizes the vehicle burden and reduces the design complexity. Secondly, the QCP-SD sends alerts to the vehicles in the risky region through highly efficient next-hop disseminators selected based on a multi-attribute cost value. Moreover, the performance of QCP-SD is evaluated through Network Simulator (NS-2). The efficiency is analyzed using the performance metrics that are duplicate packet, latency, packet loss, packet delivery ratio, secondary collision, throughput, and overhead.

Multi-Stage Estimation Algorithm for Vehicle Trajectory Tracking

This paper develops a multi-stage estimation algorithm for vehicle trajectory tracking applications. Previously designed nonlinear observers for vehicle trajectory tracking lack either the ability to handle variable velocity or have a high sensitivity to sensor noise. To overcome these shortcomings, the original model of the non-ego vehicle is translated into three separate models for speed, orientation, and position. Three stable observers are designed for these models which are all shown to be stable and robust to uncertainties. The new estimation algorithm outperforms previous high-gain and LMI-based nonlinear observers. The developed observer is useful for collision prediction and avoidance applications.

Collision Prediction of Smart Cities Using Co-simulation

Road traffic has become prominent in everyday living, impacting or disrupting services to people and daily routines. With the rise in automobile manufacturing and the frequency of vehicle crashes, human catastrophes, like fatalities, accidents, impairments, and destruction of property, are surging yearly. Vehicle collision detection has recently gained prominence in decreasing manually operated and autonomous vehicle fatalities. The concept of independent and self-driving cars relies on accurate object recognition, including pedestrians, vehicles, buildings, and other moving objects. Various object-detecting approaches have been proposed to help autonomous vehicles (AVs) achieve consistent, safe driving. Object prediction and detection have noticed numerous algorithmic changes that have improved speed and accuracy. In this study, I used a traffic dataset produced by a CARLA simulator to anticipate collisions using the Yolov7 model. I generated the dataset from a CARLA simulation bench in video sequences, manually annotated the frames, and used the deep learning algorithm Yolov7 to train them. The model predicts the collision a few seconds before it occurs in real-time. I implemented this framework to increase the safety of driving in self-driving vehicles.