Safety Of Autonomous Vehicles Research Articles

Enhanced Vision-Based Taillight Signal Recognition for Analyzing Forward Vehicle Behavior.

This study develops a vision-based technique for enhancing taillight recognition in autonomous vehicles, aimed at improving real-time decision making by analyzing the driving behaviors of vehicles ahead. The approach utilizes a convolutional 3D neural network (C3D) with feature simplification to classify taillight images into eight distinct states, adapting to various environmental conditions. The problem addressed is the variability in environmental conditions that affect the performance of vision-based systems. Our objective is to improve the accuracy and generalizability of taillight signal recognition under different conditions. The methodology involves using a C3D model to analyze video sequences, capturing both spatial and temporal features. Experimental results demonstrate a significant improvement in the model's accuracy (85.19%) and generalizability, enabling precise interpretation of preceding vehicle maneuvers. The proposed technique effectively enhances autonomous vehicle navigation and safety by ensuring reliable taillight state recognition, with potential for further improvements under nighttime and adverse weather conditions. Additionally, the system reduces latency in signal processing, ensuring faster and more reliable decision making directly on the edge devices installed within the vehicles.

LDIPRS: A novel longitudinal driving intention prior recognition technique empowered by TENG and deep learning

For autonomous vehicles, real-time and accurate longitudinal driving intention recognition is crucial as it effectively enhances driving safety and improves the driving experience. This study proposes a novel data and model hybrid-driven fine-grained longitudinal driving intention prior recognition system (LDIPRS). Firstly, the system integrates a human-pedal interaction sensor (HPIS) based on triboelectric nanogenerators for fine-grained longitudinal driving maneuver monitoring and the channel attention (CA)-enhanced convolutional neural network (CBRCNet). The HPIS, integrated into the vehicle's acceleration and brake pedals, is capable of monitoring driver foot movement information in the form of electrical signals before the vehicle responds, achieving data level advance. The collected electrical signals are fed into the CBRCNet network, which models and learns the mapping relationship between these signals and fine-grained longitudinal driving intentions, leading to model level advances. The HPIS completes the capture of longitudinal maneuver information 541 ms before the driving simulator starts to respond at the data level. At the model level, CBRCNet can achieve a recognition accuracy of 96.1 % based on partial response data (50 ms after starting response) rather than complete response data of the HPIS. Finally, our proposed LDIPRS realizes the recognition of emergency braking, rapid acceleration, normal braking, and normal acceleration in advance by 732 ms, 1035 ms, 1757 ms, and 2227 ms, respectively. This study introduces self-powered, low-cost, highly sensitive triboelectric sensors into the field of intention recognition, and combines the triboelectric sensors with deep learning algorithms to offer a promising solution to improve the safety of autonomous vehicles and the efficiency of intelligent transportation systems.

Investigation of the impacts of the deployment of autonomous vehicles on first responders

Purpose The purpose of this study is to address the significant impact AVs will have on public services and the ability of first responders to conduct their jobs safely and effectively. Autonomous vehicles (AVs) are expected to drastically change the transportation industry, and it is vital that first responders be equipped to integrate them into their occupational responsibilities. Design/methodology/approach A systematic literature review was conducted, and following a multistep exclusion process, 161 articles were selected for detailed review. The impacts of AVs on first responders were identified, classified and categorized into lists of challenges and opportunities. Based on the findings of the literature review, a SWOT (strengths, weaknesses, opportunities and threats) analysis was conducted, and stakeholder management strategies were designed. Findings Through the examination of the impacts of AVs on first responders, 17 identified challenges and opportunities were classified into the following categories: AV-related emergency response and training, perceptions and acceptance of AVs, technology development and laws and regulations. The study revealed that the optimal benefits of AVs would require stakeholders to focus more on how they interact with first responders; thus, 14 stakeholder management strategies were identified. First responders, AV manufacturers, legislators and future research paths will all benefit from this study, as it can facilitate smooth interactions between AVs and first responders. Originality/value A range of studies have been published on the safety of AVs and the public’s perceptions of this new technology; however, the integration of AVs and their interactions with first responders has been neglected. The goal of this study was to fill that research gap by providing a thorough synthesis of autonomous driving systems in the context of their interactions with first responders.

Enhancing autonomous vehicle navigation using SVM-based multi-target detection with photonic radar in complex traffic scenarios

Efficient transportation systems are essential for the development of smart cities. Autonomous vehicles and Intelligent Transportation Systems (ITS) are crucial components of such systems, contributing to safe, reliable, and sustainable transportation. They can reduce traffic congestion, improve traffic flow, and enhance road safety, thereby making urban transportation more efficient and environmentally friendly. We present an innovative combination of photonic radar technology and Support Vector Machine classification, aimed at improving multi-target detection in complex traffic scenarios. Central to our approach is the Frequency-Modulated Continuous-Wave photonic radar, augmented with spatial multiplexing, enabling the identification of multiple targets in various environmental conditions, including challenging weather. Notably, our system achieves an impressive range resolution of 7 cm, even under adverse weather conditions, utilizing an operating bandwidth of 4 GHz. This feature is particularly crucial for precise detection and classification in dynamic traffic environments. The radar system's low power requirement and compact design enhance its suitability for deployment in autonomous vehicles. Through comprehensive numerical simulations, our system demonstrated its capability to accurately detect targets at varying distances and movement states, achieving classification accuracies of 75% for stationary and 33% for moving targets. This research substantially contributes to ITS by offering a sophisticated solution for obstacle detection and classification, thereby improving the safety and efficiency of autonomous vehicles navigating through urban environments.

A method based on Vision Transformer and multiple image information for vehicle lane-changing recognition in mixed traffic and connected environment

ABSTRACT To enhance the safety of autonomous vehicles in mixed traffic and connected environment, it is crucial to recognize the lane-changing intentions (LCIs) of human-driven vehicles for autonomous vehicles. This paper presents a novel method for LCI recognition, which extracts features from the driving state and relative motion of the target vehicle and its neighbors. The method applies short-time Fourier transform, Gramian angular summation field, and Gramian angular difference field to the time-series data, and generates three grayscale images, which are merged into one information fusion image (IFI) by image processing techniques. The IFIs are then classified into three categories: lane keeping, lane-changing left, and lane-changing right, using the Vision Transformer model with transfer learning to speed up convergence and reduce training cost. The experimental results demonstrate that the proposed method outperforms the traditional methods, achieving an accuracy of 95.65% for recognizing LCI 3s before the lane change point.

Vision-Based Algorithm for Precise Traffic Sign and Lane Line Matching in Multi-Lane Scenarios

With the rapid development of intelligent transportation systems, lane detection and traffic sign recognition have become critical technologies for achieving full autonomous driving. These technologies offer crucial real-time insights into road conditions, with their precision and resilience being paramount to the safety and dependability of autonomous vehicles. This paper introduces an innovative method for detecting and recognizing multi-lane lines and intersection stop lines using computer vision technology, which is integrated with traffic signs. In the image preprocessing phase, the Sobel edge detection algorithm and weighted filtering are employed to eliminate noise and interference information in the image. For multi-lane lines and intersection stop lines, detection and recognition are implemented using a multi-directional and unilateral sliding window search, as well as polynomial fitting methods, from a bird’s-eye view. This approach enables the determination of both the lateral and longitudinal positioning on the current road, as well as the sequencing of the lane number for each lane. This paper utilizes convolutional neural networks to recognize multi-lane traffic signs. The required dataset of multi-lane traffic signs is created following specific experimental parameters, and the YOLO single-stage target detection algorithm is used for training the weights. In consideration of the impact of inadequate lighting conditions, the V channel within the HSV color space is employed to assess the intensity of light, and the SSR algorithm is utilized to process images that fail to meet the threshold criteria. In the detection and recognition stage, each lane sign on the traffic signal is identified and then matched with the corresponding lane on the ground. Finally, a visual module joint experiment is conducted to verify the effectiveness of the algorithm.

Enhancing autonomous driving safety: A robust traffic sign detection and recognition model TSD-YOLO

As autonomous driving technology rapidly advances, Traffic Sign Detection and Recognition (TSDR) has become pivotal in ensuring the safety and regulatory compliance of autonomous vehicles. Despite progress, existing technologies struggle under challenging conditions such as adverse weather and complex roadway environments. To overcome these obstacles, we introduce a novel model, TSD-YOLO, which leverages Mamba and YOLO technologies to enhance the accuracy and robustness of traffic sign detection. Our innovative YOLO-MAM dual-branch module merges convolutional layer-based local feature extraction with the long-distance dependency capabilities of the State Space Models (SSMs). We conducted experimental validations using the Tsinghua-Tencent 100K (TT-100K) dataset and the Mapillary Traffic Sign Detection (MTSD) dataset, demonstrating our model’s efficacy across various datasets. Furthermore, cross-dataset validations affirm the model’s exceptional generalization and robustness across diverse environments. This study not only bolsters traffic sign detection and recognition in autonomous driving systems but also paves the way for future advancements in autonomous driving technology.

Wavelet-powered hierarchical frequency filtering framework for autonomous vehicle sensors fault diagnosis and correction under open environments

The reliable operation of autonomous vehicles heavily depends on the functionality of numerous sensors that facilitate environmental perception and vehicle control. However, sensor malfunctions can significantly undermine the dependability and safety of autonomous vehicles. Despite the significant advancements made in sensor fault diagnosis through deep learning, the technology's susceptibility to noise and limited frequency analysis capacity renders it less effective in addressing frequency aliasing and noise interference issues. In addition, repairing incorrect sensor data holds greater significance for autonomous vehicles than merely identifying sensor failures. To this end, this paper proposes a wavelet-powered hierarchical frequency filtering framework (WavePHF) for autonomous vehicle sensors' fault diagnosis and correction. First, this study designs a multi-sensor fault identification and correction joint architecture to detect sensor faults and repair fault signals in time. Second, this study integrates the discrete wavelet transform (DWT) algorithm into the deep model to significantly enhance the model's frequency analysis capabilities and suppress frequency aliasing and noise interference. In particular, DWT is deeply integrated into the convolutional modules' parameter updating and optimization process, thereby facilitating mutual learning and promotion. Third, this study proposes an adaptive frequency weighting block for low-frequency features and an adaptive frequency discarding block for high-frequency features to achieve hierarchical frequency filtering in a coarse-to-fine manner and efficiently identify fault categories and restore pure signals. The WavePHF is evaluated on a real autonomous driving dataset. Experiments show that WavePHF has excellent autonomous vehicle sensors fault diagnosis and correction capabilities and competitive performance in different urban scenarios.

Dependent Hidden Markov Model for pedestrian intention prediction: considering Multivariate Interaction Force

Accurately recognizing and predicting pedestrian intentions is crucial for autonomous vehicle safety. However, existing prediction models often fail to comprehensively consider interactions between various traffic elements, resulting in suboptimal accuracy and robustness, especially in complex environments. To address this, we propose a pedestrian intention prediction model combining the Multivariate Interaction Force (MIF) model and a Dependent Hidden Markov Model (DE-HMM) for unsignalized midblock crossings. The MIF model captures dynamic interactions among pedestrians, vehicles, and the environment, while DE-HMM uses MIF data and pedestrian head orientation for predictions. Our model achieves 91.5% accuracy in recognizing crossing intentions, and 88.7% and 85.1% accuracy for predictions 0.5s and 1s ahead, respectively, outperforming current mainstream models and demonstrating strong robustness in special scenarios.

Research on a Path Tracking Control Strategy for Autonomous Vehicles Based on State Parameter Identification

With the rapid development of autonomous driving technology, estimating and controlling key vehicle state parameters under complex road conditions have become critical challenges. This study combines Unscented Kalman Filtering (UKF) and Sliding Mode Control (SMC) methods to propose an integrated control model for achieving more efficient control. First, a three-degrees-of-freedom vehicle dynamics model based on the Dugoff tire model is constructed to accurately estimate key vehicle state parameters. Next, UKF is used to estimate road friction coefficients and key vehicle state parameters, and its performance is compared with Extended Kalman Filtering (EKF) under various conditions. The results show the superiority of UKF in identifying road friction coefficients. Based on SMC theory, a sliding surface is designed, and the functional relationship between state variables and control variables is derived to establish the corresponding control model. Joint simulations using Carsim and Simulink under different conditions validate the real-time performance and effectiveness of the designed UKF-SMC integrated control strategy in the presence of external disturbances and system uncertainties. Simulation results indicate that this strategy effectively enhances the overall performance and safety of autonomous vehicles, providing an accurate real-time solution capable of handling complex and variable road conditions. The proposed UKF-SMC integrated control strategy not only proves its theoretical superiority but also demonstrates promising practical applications in simulation experiments. This study provides reliable technical support for the development of autonomous driving technology under complex road conditions.

INTEGRATION OF MACHINE LEARNING-BASED DETECTION SYSTEMS INTO AUTONOMOUS VEHICLES

The incorporation of machine learning (ML) driven detecting systems into autonomous vehicles (AVs) signifies a revolutionary advancement in contemporary transportation. The present study investigates the use of ML systems, namely in the domains of traffic sign identification, pedestrian detection, and obstacle resolution. The work deals with a review of the technical advances in neural networks concerning their capability for real-time data processing for accurate and reliable navigation. Further, this paper points out that the challenges to be faced during the implementation of these systems include requirements for data management, high-performance computing, and exploitation of cloud technologies for scalable solutions. The results indicate that machine learningbased detection strategies greatly enhance autonomous vehicle performance and safety, while emphasising persistent issues concerning security and legal frameworks. This paper highlights the significance of ongoing technological advancements in machine learning and its influence on the development of autonomous driving.

The Effect of Directional Tactile Memory of the Back of the User on Reaction Time and Accuracy

Tactile memory is the cognitive process of storing and recalling information that has been perceived through the sense of touch. Directional tactile memory involves the encoding and retrieval of sensory data associated with a tactile experience, allowing individuals to remember and recognize directional information encoded through the sense of touch. A new method for providing directional tactile feedback, at the back of the user, has been developed to investigate the efficacy of directional tactile memory, its decay over time, and its impact during a concurrent cognitive task. Two experiments were presented. In the first experiment, tactile memory deterioration, with a visual or a tactile cue, was tested with different action-cue latencies (10 s and 20 s). In the second experiment, we considered tactile memory deterioration when there was an increased cognitive load as the participants played Tetris. Forty volunteers participated in the two experiments using purpose-built tactile seats with nine motors controlled by an Arduino. The performance data (error and reaction times) were analyzed statistically, and a NASA task load index (NASA-TLX) questionnaire was administered to measure the subjective workload after each of the two experiments. The findings highlighted that the directional tactile memory of the back can guide individuals to the correct point on the screen and that it can be maintained for at least 20 s. There was no statistically significant difference in the number of errors or reaction time with a visual or tactile action cue. However, being involved in a concurrent cognitive task (playing Tetris) adversely affected the reaction time, the number of errors, and the directional tactile memory, which degraded as the time between the directional cue and the action cue increased. Participants perceived the performance while playing Tetris as significantly more mentally and perceptually demanding, requiring more mental and physical effort and being more frustrating. These trials revealed a new potential for a human–machine interface system, leveraging directional tactile memory, which might be utilized to increase the safety of autonomous vehicles.

Balancing Safety and Decision-Making Algorithms in AI-Driven Autonomous Vehicles

Balancing Safety and Decision-Making Algorithms in AI-Driven Autonomous Vehicles

Optimal hybrid strategy in adaptive cruise control system for enhanced autonomous vehicle stability and safety

The adaptive cruise control system relies heavily on the vehicle's longitudinal control algorithm. Traditional longitudinal control algorithms typically depend on precise vehicle dynamic modeling or complex controller parameter calibrations. In the research, an optimal hybrid strategy for the control of basic adaptive cruise control system is proposed. The proposed hybrid strategy is the combination of both the Model Predictive Control (MPC) and Coati Optimization Algorithm (COA) and hence it is named as MPCCOA strategy. The fundamental goal of the proposed work is that a vehicle should maintain its current velocity when it approaches another vehicle that is driving at a steady velocity. The inputs for the proposed strategy are vehicle speed and acceleration. The inter-vehicle distance error, velocity error and acceleration are regulated based on the upper and lower limit of acceleration command with the help of proposed strategy. For accomplishing the required steadiness with the limitations in the given input and actual velocity with constant previous vehicle velocity, the proposed controller is utilized. The proposed controller is used to control the traction force of the host vehicle such that the vehicle speed follows the optimal speed trajectory as good as possible while ensuring constraints like distance to a preceding vehicle or speed limits. The performance of the proposed is implemented in the MATLAB platform and compared with existing approaches. The proposed technique demonstrates superior performance metrics, achieving the collision rate, rate of reaching maximum deceleration and ratio of exceeding comfortable deceleration or jerk are 1.27 %, 3.131 % and 9.052 %, respectively.

Onward and Autonomously: Expanding the Horizon of Image Segmentation for Self-Driving Cars through Machine Learning

Autonomous navigation is the leading technology in current era, in this intelligent traffic light, sign detection, ADAS and obstacle detections were playing major role. Image segmentation is the process of dividing an image into different regions, or semantic classes. This is a challenging problem in autonomous vehicle technology because it requires the vehicle to be able to understand its surroundings to safely navigate. The major challenges in this platform are the accuracy and efficiency of model performance. The proposed method in the abstract uses a convolutional neural network (CNN) to perform image segmentation. CNNs are a type of deep learning model that is well-suited for image processing tasks. The CNN in this paper was trained on a local city dataset, and it was able to achieve a mean intersection over union (IoU) of 73%. IoU is a measure of how well the segmentation results match the ground truth labels. A score of 100% indicates that the segmentation is perfect, while a score of 0% indicates that the segmentation is completely wrong. This means that the method can segment images at a very fast rate, which is important for autonomous vehicles that need to make real-time decisions. Overall, the proposed method is a promising approach for image segmentation in autonomous vehicles. It can achieve high accuracy and speed, and it is easy to implement using Python. The proposed method attains an accuracy of 98.34 %, a Sensitivity of 97.26 % and a sensitivity of 96.37 % had been attained. The method could be used to improve the safety and efficiency of autonomous vehicles by enabling them to better understand their surroundings.

Autonomous Vehicle Safety through the SIFT Method: A Conceptual Analysis

This study aims to provide a conceptual analysis of the dynamic transformations occurring in an autonomous vehicle (AV), placing a specific emphasis on the safety implications for pedestrians and passengers. AV, also known as self-driving automobiles, are positioned as potential disruptors in the contemporary transportation landscape, offering heightened safety and improved traffic efficiency. Despite these promises, the intricate nature of road scenarios and the looming specter of misinformation pose challenges that can compromise the efficacy of AV decision-making. A crucial aspect of the proposed verification process is the incorporation of the stop, investigate the source, find better coverage, trace claims, quotes, and media to the original context (SIFT) method. The SIFT method, originally designed to combat misinformation, emerges as a valuable mechanism for enhancing AV safety by ensuring the accuracy and reliability of information influencing autonomous decision-making processes.

BIPOOLNET: An advanced UNet architecture for enhanced lane detection in autonomous vehicles

In the rapidly evolving landscape of autonomous vehicle technology, the imperative to bolster safety within smart urban ecosystems has never been more critical. This endeavor requires the deployment of advanced detection systems capable of navigating the intricacies of pedestrian, near-vehicle, and lane detection challenges, with a particular focus on the nuanced requirements of curved lane navigation – a domain where traditional AI models exhibit notable deficiencies. This paper introduces BIPOOLNET, an innovative encoder-decoder neural architecture, ingeniously augmented with a feature pyramid to facilitate the precise delineation of curved lane geometries. BIPOOLNET integrates max pooling and average pooling to extract critical features and mitigate the complexity of the feature map, redefining the benchmarks for lane detection technology. Rigorous evaluation using the TuSimple dataset underscores BIPOOLNET’s exemplary performance, evidenced by an unprecedented accuracy rate of 98.45%, an F1-score of 98.17%, and notably minimal false positive (1.84%) and false negative (1.09%) rates. These findings not only affirm BIPOOLNET’s supremacy over extant models but also signal a paradigm shift in enhancing the safety and navigational precision of autonomous vehicles, offering a scalable, robust solution to the multifaceted challenges posed by real-world driving dynamics.

Path-tracking control at the limits of handling of a prototype over-actuated autonomous vehicle

Considering the vehicle dynamics at the limits of handling is vital to improve the performance and safety of autonomous vehicles especially in extreme situations. This paper presents the development of a path-tracking controller for an over-actuated autonomous vehicle. The vehicle is an electric prototype equipped with torque vectoring and four-wheel steering, which enable enhanced control of vehicle dynamics. A model predictive controller is proposed taking into account the nonlinearities in vehicle dynamics at the limits of handling as well as the crucial actuator constraints. The controller is examined in both high-fidelity simulation and practical testing to validate the vehicle's handling performance. Both the simulation and testing results illustrate that the over-actuation topology can enhance the handling performance as well as vehicle stability at conditions close to the limits of handling. With additional references such as side slip angle, the vehicle's attitude under such extreme condition can also be manipulated. The testing also demonstrates the real-time capability of the controller. Further testing has been done to confirm that side slip angle reference plays an important role in path-tracking control at the limits of handling, and to push the vehicle to the friction limits.

GPT-4V Takes the Wheel: Promises and Challenges for Pedestrian Behavior Prediction

Predicting pedestrian behavior is the key to ensure safety and reliability of autonomous vehicles. While deep learning methods have been promising by learning from annotated video frame sequences, they often fail to fully grasp the dynamic interactions between pedestrians and traffic, crucial for accurate predictions. These models also lack nuanced common sense reasoning. Moreover, the manual annotation of datasets for these models is expensive and challenging to adapt to new situations. The advent of Vision Language Models (VLMs) introduces promising alternatives to these issues, thanks to their advanced visual and causal reasoning skills. To our knowledge, this research is the first to conduct both quantitative and qualitative evaluations of VLMs in the context of pedestrian behavior prediction for autonomous driving. We evaluate GPT-4V(ision) on publicly available pedestrian datasets: JAAD and WiDEVIEW. Our quantitative analysis focuses on GPT-4V's ability to predict pedestrian behavior in current and future frames. The model achieves a 57% accuracy in a zero-shot manner, which, while impressive, is still behind the state-of-the-art domain-specific models (70%) in predicting pedestrian crossing actions. Qualitatively, GPT-4V shows an impressive ability to process and interpret complex traffic scenarios, differentiate between various pedestrian behaviors, and detect and analyze groups. However, it faces challenges, such as difficulty in detecting smaller pedestrians and assessing the relative motion between pedestrians and the ego vehicle.

Trusted Perception Method for Traffic Signs That Are Physically Attacked

Traffic sign recognition is a crucial method by which autonomous driving systems acquire road information, and is predominantly based on deep neural networks (DNNs). However, the recognition results of DNNs are not always trustworthy for traffic signs subject to abnormal disturbance. Recently, the phenomenon of adversarial examples successfully deceiving DNNs has garnered considerable attention. Because DNN-based computer vision techniques are becoming increasingly prevalent in traffic scenarios, the misclassification of attacked traffic signs by DNN classifiers poses serious safety hazards. Although numerous methods have been proposed for crafting physical adversarial examples that are robust in the real world, most existing defense approaches focus on digital attacks, which necessitate the adversary infiltrating the embedded system; thus, it becomes challenging to obtain results. A reliable approach for defending against physical adversarial traffic signs enables autonomous vehicles to achieve trusted perception of traffic signs. In this paper, we present a deep image prior-based pipeline to defend against robust adversarial traffic signs in the real world, an approach that circumvents the need for prior data sets during training. Our approach protects the safety of autonomous vehicles by performing image reconstruction of captured traffic sign images. The genuine traffic sign class can be inferred by leveraging the consistency of the victim classifier’s decision results for reconstructed images at different stages. Additionally, we evaluate the efficacy of our defense pipeline for detecting other potential types of physical adversarial traffic signs that may exist in the real world, thus demonstrating the generalizability of our approach.