Collision Prediction Research Articles

위험 시나리오에 대한 트랜스포머 네트워크 기반 자율주행차량의 충돌 및 궤적 예측 알고리즘 개발

위험 시나리오에 대한 트랜스포머 네트워크 기반 자율주행차량의 충돌 및 궤적 예측 알고리즘 개발

Real-time accident risk identification for freeway weaving segments based on video analytics

Accurately identifying the risk of traffic accidents in real time for specific road sections is crucial for implementing proactive traffic control strategies. The existing studies have focused on real-time prediction of collisions between two vehicles, identification of accident risks on road segments relies heavily on historical accident data. This study proposes a framework for identifying road section accident risks at the granular level by utilizing traffic conflicts as an intermediary. In this framework, vehicles are detected and tracked in real time and the pixel coordinates are then efficiently converted into world coordinates based on the camera projection principle. A binary logit model is constructed to identify road section accident risks using carefully selected macroscopic traffic flow parameters. The proposed framework was validated on two one-sided ramp weaves in Jinan, China, achieving an accuracy of 76.2 %. The findings will serve as a foundation for traffic safety monitoring and proactive traffic control.

A systematic review of factors, data sources, and prediction techniques for earlier prediction of traffic collision using AI and machine Learning

A systematic review of factors, data sources, and prediction techniques for earlier prediction of traffic collision using AI and machine Learning

Autonomous navigation and collision prediction of port channel based on computer vision and lidar.

This study aims to enhance the safety and efficiency of port navigation by reducing ship collision accidents, minimizing environmental risks, and optimizing waterways to increase port throughput. Initially, a three-dimensional map of the port's waterway, including data on water depth, rocks, and obstacles, is generated through laser radar scanning. Visual perception technology is adopted to process and identify the data for environmental awareness. Single Shot MultiBox Detector (SSD) is utilized to position ships and obstacles, while point cloud data create a comprehensive three-dimensional map. In order to improve the optimal navigation approach of the Rapidly-Exploring Random Tree (RRT), an artificial potential field method is employed. Additionally, the collision prediction model utilizes K-Means clustering to enhance the Faster R-CNN algorithm for predicting the paths of other ships and obstacles. The results indicate that the RRT enhanced by the artificial potential field method reduces the average path length (from 500 to 430m), average time consumption (from 30 to 22s), and maximum collision risk (from 15 to 8%). Moreover, the accuracy, recall rate, and F1 score of the K-Means + Faster R-CNN collision prediction model reach 92%, 88%, and 90%, respectively, outperforming other models. Overall, these findings underscore the substantial advantages of the proposed enhanced algorithm in autonomous navigation and collision prediction in port waterways.

Vehicle-to-Cyclist Collision Prediction Models by Applying Machine Learning Techniques to Virtual Reality Bicycle Simulator Data

The study of vulnerable road users (VRUs) behavior is key to designing and optimizing driving assistance systems, such as the autonomous emergency braking (AEB) system. These kinds of devices could help lower the VRU accident rate, which is of particular interest to cyclists, who are the subject of this research. To better understand cyclists’ reaction patterns in frequently occurring collision scenarios in urban environments, this paper focuses on developing a virtual reality (VR) simulator for cyclists (VRBikeSim) that incorporates eye-tracking functionality. The braking and steering systems were calibrated by means of on-track tests with a sensorized bicycle in order to improve the accuracy of the bicycle virtual model. From the data obtained in the virtual tests, a battery of predictive models was built using supervised machine learning classifiers. All of them exhibited an accuracy higher than 85%, especially the K-Nearest Neighbors model. This model allowed us to obtain the best balance between the prediction of avoidance and collision cases, as well as enabling computationally lower times to be incorporated into the decision-making algorithm of an AEB system.

A real-time digital twin for active safety in an aircraft hangar

The aerospace industry prioritises safety protocols to prevent accidents that can result in injuries, fatalities, or aircraft damage. One of the potential hazards that can occur while manoeuvring aircraft in and out of a hangar is collisions with other aircraft or buildings, which can lead to operational disruption and costly repairs. To tackle this issue, we have developed the Smart Hangar project, which aims to alert personnel of increased risks and prevent incidents from happening. The Smart Hangar project uses computer vision, LiDAR, and ultra-wideband sensors to track all objects and individuals within the hangar space. These data inputs are combined to form a real-time 3D Digital Twin (DT) of the hangar environment. The Active Safety system then uses the DT to perform real-time path planning, collision prediction, and safety alerts for tow truck drivers and hangar personnel. This paper provides a detailed overview of the system architecture, including the technologies used, and highlights the system’s performance. By implementing this system, we aim to reduce the risk of accidents in the aerospace industry and increase safety for all personnel involved. Additionally, we identify future research directions for the Smart Hangar project.

Machine Learning-based Identification of Contaminated Images in Light Curve Data Preprocessing

Attitude is one of the crucial parameters for space objects and plays a vital role in collision prediction and debris removal. Analyzing light curves to determine attitude is the most commonly used method. In photometric observations, outliers may exist in the obtained light curves due to various reasons. Therefore, preprocessing is required to remove these outliers to obtain high quality light curves. Through statistical analysis, the reasons leading to outliers can be categorized into two main types: first, the brightness of the object significantly increases due to the passage of a star nearby, referred to as “stellar contamination,” and second, the brightness markedly decreases due to cloudy cover, referred to as “cloudy contamination.” The traditional approach of manually inspecting images for contamination is time-consuming and labor-intensive. However, we propose the utilization of machine learning methods as a substitute. Convolutional Neural Networks and SVMs are employed to identify cases of stellar contamination and cloudy contamination, achieving F1 scores of 1.00 and 0.98 on a test set, respectively. We also explore other machine learning methods such as ResNet-18 and Light Gradient Boosting Machine, then conduct comparative analyses of the results.

Selection of Lightweight CNN Models with Limited Computing Resources for Drone Collision Prediction

The Collision Avoidance System (CAS) is a safety system created to identify and prevent collisions, primarily on drones. The CAS comprises three processes: detection, prediction, and action. The predictive process is crucial as it determines whether a collision will occur, making it the core component of the system. Most drones are equipped with cameras. A visual-based prediction involves the use of a convolutional neural network (CNN). The CNN operates by autonomously learning and extracting hierarchical characteristics from input data through convolution, pooling, and fully connected layers. Currently, there are CNN models called pretrained models that are ready to use. However, not all pretrained models are suitable for compatibility with drones as they possess computational constraints. Our objective is to establish a suitable model selection from a variety of pre-trained CNN models with lightweight architectures. The transfer learning technique is applied to customize these models with the ColANet dataset. Subsequently, we evaluate these models regarding their accuracy, model size, inference time, and power consumption. Finally, the selected model is deployed in real time on a Raspberry Pi 3B+ with data input from a DJI Tello drone camera, and the prediction performance is evaluated.

Stereotactic Optimized Automated Radiotherapy (SOAR): a novel automated planning solution for multi-metastatic SRS compared to HyperArc™

Objective. Automated Stereotactic Radiosurgery (SRS) planning solutions improve clinical efficiency and reduce treatment plan variability. Available commercial solutions employ a template-based strategy that may not be optimal for all SRS patients. This study compares a novel beam angle optimized Volumetric Modulated Arc Therapy (VMAT) planning solution for multi-metastatic SRS to the commercial solution HyperArc. Approach. Stereotactic Optimized Automated Radiotherapy (SOAR) performs automated plan creation by combining collision prediction, beam angle optimization, and dose optimization to produce individualized high-quality SRS plans using Eclipse Scripting. In this retrospective study 50 patients were planned using SOAR and HyperArc. Assessed dose metrics included the Conformity Index (CI), Gradient Index (GI), and doses to organs-at-risk. Complexity metrics evaluated the modulation, gantry speed, and dose rate complexity. Plan dosimetric quality, and complexity were compared using double-sided Wilcoxon signed rank tests (α = 0.05) adjusted for multiple comparisons. Main Results. The median target CI was 0.82 with SOAR and 0.79 with HyperArc (p < .001). Median GI was 1.85 for SOAR and 1.68 for HyperArc (p < .001). The median V12Gy normal brain volume for SOAR and HyperArc were 7.76 cm3 and 7.47 cm3 respectively. Median doses to the eyes, lens, optic nerves, and optic chiasm were statistically significant favoring SOAR. The SOAR algorithm scored lower for all complexity metrics assessed. Significance. In-house developed automated planning solutions are a viable alternative to commercial solutions. SOAR designs high-quality patient-specific SRS plans with a greater degree of versatility than template-based methods.

Using explainable AI for enhanced understanding of winter road safety: insights with support vector machines and SHAP

This study investigates the utility of machine learning (ML) in understanding and mitigating winter road risks. Despite their capability in managing complex data structures, ML models often lack interpretability. We address this issue by integrating Shapley Additive exPlanations (SHAP) with a support vector machine (SVM) model. Utilizing a comprehensive dataset of 231 snowstorm events collected in the city of Edmonton across two winter seasons, the SVM model achieved an accuracy rate of 87.2%. Following model development, a SHAP summary plot was employed to identify the contribution of individual features to collision predictions—an insight not achievable through ML alone. Next, SHAP waterfall plots were used to assess the reliability of individual predictions. The findings enhanced our understanding of the complex SVM model and provided greater insights into the diverse factors affecting winter road safety.

Advancements in Machine Learning and Data Mining Techniques for Collision Prediction and Hazard Detection in Internet of Vehicles

Advancements in Machine Learning and Data Mining Techniques for Collision Prediction and Hazard Detection in Internet of Vehicles

Dynamic short-term crash analysis and prediction at toll plazas for proactive safety management

Toll plazas are commonly recognized as bottlenecks on toll roads, where vehicles are prone to crashes. However, there has been a lack of research analyzing and predicting dynamic short-term crash risk specifically at toll plazas. This study utilizes traffic, geometric, and weather data to analyze and predict dynamic short-term collision occurrence probability at mainline toll plazas. A random-effects logit regression model is employed to identify crash precursors and assess their impacts on the probability of crash occurrence at toll plazas. Meanwhile, a Long Short-Term Memory Convolutional Neural Network (LSTM-CNN) network is applied for crash prediction. The results of random-effects logit regression model indicate that the flow standard deviation of downstream, upstream occupancy, speed difference and occupancy difference between upstream and downstream positively influence the probability of crash occurrence. Conversely, an increase in the proportion of ETC lanes negatively impacts the probability of crash occurrence. Additionally, there appears a higher likelihood of crashes occurring during summer at toll plaza area. Furthermore, to address the issue of data imbalance, Synthetic Minority Oversampling Techniques (SMOTE) and class weight methods were employed. Stacked Sparse AutoEncoder-Long Short-Term Memory (SSAE-LSTM) and CatBoost were developed and their performance was compared with the proposed model. The results demonstrated that the LSTM-CNN model outperformed the other models in terms of the Area Under the Curve (AUC) values and the true positive rate. The findings of this study can assist engineers in selecting suitable traffic control strategies to improve traffic safety in toll plaza areas. Moreover, the developed collision prediction model can be incorporated into a real-time safety management system to proactively prevent traffic crash.

Collision Prediction in an Integrated Framework of Scenario-Based and Data-Driven Approaches

Collision Prediction in an Integrated Framework of Scenario-Based and Data-Driven Approaches

A Novel Hybrid Model Based on Rule Learning and Dilated Convolution for Vehicle Collision Prediction

A Novel Hybrid Model Based on Rule Learning and Dilated Convolution for Vehicle Collision Prediction

Image Recognition Technology Applied to the Design of Mobile Platform for Warehouse Logistics Robots

Abstract This paper first studies the processing flow of image processing technology that preprocesses the image and adopts the method of polygonal approximation to identify the shape and localize the moving target. Then, the mobile platform of the warehouse logistics robot is designed. Then, the vision system of the robot was designed using image recognition technology to realize obstacle collision prediction and route planning. Finally, the robot’s localization and grasping abilities, trajectory following performance, and semantic segmentation abilities are analyzed using comparative experiments. The successful localization and grasping rates of the warehouse robots are all higher than 93%, and the trajectory following the straight line road section is better, with a maximum error of less than 21 mm. The mIoU of this paper’s method on the Cityscapes dataset is 78.85%, MPA is 86.05%, and PA is 96.89%, with good image segmentation performance. This study is of great significance for the development of the intelligent logistics field.

Uncertainty-aware visually-attentive navigation using deep neural networks

Autonomous navigation and information gathering in challenging environments are demanding since the robot’s sensors may be susceptible to non-negligible noise, its localization and mapping may be subject to significant uncertainty and drift, and performing collision-checking or evaluating utility functions using a map often requires high computational costs. We propose a learning-based method to efficiently tackle this problem without relying on a map of the environment or the robot’s position. Our method utilizes a Collision Prediction Network (CPN) for predicting the collision scores of a set of action sequences, and an Information gain Prediction Network (IPN) for estimating their associated information gain. Both networks assume access to a) the depth image (CPN) or the depth image and the detection mask from any visual method (IPN), b) the robot’s partial state (including its linear velocities, z-axis angular velocity, and roll/pitch angles), and c) a library of action sequences. Specifically, the CPN accounts for the estimation uncertainty of the robot’s partial state and the neural network’s epistemic uncertainty by using the Unscented Transform and an ensemble of neural networks. The outputs of the networks are combined with a goal vector to identify the next-best-action sequence. Simulation studies demonstrate the method’s robustness against noisy robot velocity estimates and depth images, alongside its advantages compared to state-of-the-art methods and baselines in (visually-attentive) navigation tasks. Lastly, multiple real-world experiments are presented, including safe flights at 2.5 m/s in a cluttered corridor, and missions inside a dense forest alongside visually-attentive navigation in industrial and university buildings.

Online Learning for Dynamic Impending Collision Prediction using FMCW Radar

Radar collision prediction systems can play a crucial role in safety critical applications, such as autonomous vehicles and smart helmets for contact sports, by predicting an impending collision just before it will occur. Collision prediction algorithms use the velocity and range measurements provided by radar to calculate time to collision. However, radar measurements used in such systems contain significant clutter, noise, and inaccuracies which hamper reliability. Existing solutions to reduce clutter are based on static filtering methods. In this paper, we present a deep learning approach using frequency modulated continuous wave (FMCW) radar and inertial sensing that learns the environmental and user-specific conditions that lead to future collisions. We present a process of converting raw radar samples to range-Doppler matrices (RDMs) and then training a deep convolutional neural network that outputs predictions (impending collision vs. none) for any measured RDM. The system is retrained to work in dynamically changing environments and maintain prediction accuracy. We demonstrate the effectiveness of our approach of using the information from radar data to predict impending collisions in real-time via real-world experiments, and show that our method achieves an F1-score of 0.91 and outperforms a traditional approach in accuracy and adaptability.

Aviation Collision Risk Warning for Non-Tower Airport Based on Trajectory Prediction

With the development of the civil aviation industry, the throughput of airports continues to increase, including non-tower airports (small airports without the ability to actively communicate with aircraft). These non-tower civil airports can handle small aircraft such as personal aircraft well and lay a solid foundation for future automated aircraft. However, there is a high collision risk in taking off or landing because the probability of avoiding risks highly relies on the pilot’s flight proficiency and experience. To address this problem, we studied the collision risk of aircraft in non-tower airports and proposed a collision risk warning method based on trajectory prediction. The proposed method obtains the probability of future collisions by learning the flight patterns of all aircraft in non-tower airports, the impact of environmental wind speed on pilot behavior, and the interaction patterns between aircraft. At the same time, we conducted in-depth optimization of the probability integration calculation process under a conditional threshold to reduce the time complexity greatly. The simulation experiment results on non-tower airports show that the designed method for collision risk warning can effectively address the aircraft collision prediction problem in non-tower airports, and outperform other benchmark learning methods. It has significant guiding implications for improving airport intelligence and aircraft automation in the future.

Dynamic smooth and stable obstacle avoidance for unmanned aerial vehicle based on collision prediction

AbstractAutonomous dynamic obstacle avoidance of unmanned aerial vehicles (UAVs) based on collision prediction is critical for stable flight missions. UAVs must deal with high‐speed and nonlinearly moving obstacles that obey nonlinear dynamics, which are one of the numerous objects affecting stable flight. To satisfy the requirements of smooth flight and attitude stability, a novel dynamic smooth obstacle avoidance (QVA) method based on obstacle trajectory prediction is proposed. To predict the dynamic obstacle trajectories, a trajectory prediction (QLTP) method using a quasi‐linear parameter varying representations is proposed. The proposed QVA integrates the QLTP approach, velocity obstacle (VO) approach, and artificial potential field (APF) methods. The QVA detects an imminent collision based on the QLTP method, and then replans the UAV path based on the VO method at the time of the predicted collision. The UAV tracks planned waypoints for collision avoidance. To ensure the flight safety of the UAV, a virtual APF is constructed with waypoints as local targets and obstacles. The simulation results show that the proposed method performs better than the improved APF and VO methods in terms of the smoothness of the obstacle avoidance path and the stability of the UAV attitude.

Age-Related Changes in the Utilization of Visual Information for Collision Prediction: A Study Using an Affordance-Based Model

ABSTRACT The ability to predict collisions with moving objects deteriorates with aging. We followed the affordance-based model to identify optical variables that older adults had difficulty using for collision prediction. We reproduced a modified version of the interception task used in Steinmetz (Steinmetz, Layton, Powell, & Fajen, 2020, “Affordance-based versus current – future accounts of choosing whether to pursue or abandon the chase of a moving target,” Journal of Vision, 20(3), 8) in a virtual reality (VR) environment and newly introduced perturbation for each of three optical variables (vertical and horizontal expansions of a moving object and the bearing angle produced between participants and a moving object). We expected that perturbation would negatively affect the performance only for those who rely on the optical variable to perform the interception task effectively. We tested 18 older and 15 younger adults and showed that older participants were not negatively affected by the perturbation for the vertical and horizontal expansion of a moving object, while they showed decreased performance when the perturbation was introduced with a bearing angle. These findings suggest that predicting collisions with moving objects deteriorates with aging because the perception of object expansion is impaired with aging.