Pre-crash Information Research Articles

Understanding the role of driver behaviors and performance in safety-critical events: Application of machine learning

Human factors contribute in some way to almost 93% of road crashes. Because the unavailability of good pre-crash data, the contribution of human factors to safety-critical events (SCEs) and the prediction of crashes using real-world data is lightly researched. This study provides predictive accuracy by harnessing unique real-world naturalistic driving study (NDS) data, which includes dynamic pre-crash information about driving behavior and performance. After cleaning and preprocessing, a final subsample (N = 9,237) was used and split into training and test samples. For consistent comparison of variables’ importance in statistical and machine learning (ML) models, the dominance analysis uncovered the most important predictors used by the ordered Probit model. Next, three non-parametric supervised ML methods, because promising prediction performance and cost-effectiveness, including Naïve Bayes, K-Nearest Neighbors, and Gradient Boosting Tree (GBT) were used. The overall out-of-sample prediction accuracy for the ordered Probit model was 85.75% which was lower than all three ML methods. The GBT showed the highest (91.23%) out-of-sample prediction accuracy. The availability of pre-crash naturalistic data helps significantly improve the prediction accuracy of SCEs as cumulative importance for all available human factors in the GBT classifier was 94%. For practical applications, refer to the article.

Injury Risk Assessment and Interpretation for Roadway Crashes Based on Pre-Crash Indicators and Machine Learning Methods

Pre-crash injury risk (IR) assessment is essential for guiding efforts toward active vehicle safety. This work aims to conduct crash severity assessment using pre-crash information and establish the intrinsic mechanism of IR with proper interpretation methods. The impulse–momentum theory is used to propose novel a priori formulations of several severity indicators, including velocity change (ΔV), energy equivalent speed (EES), crash momentum index (CMI), and crash severity index (CSI). Six IR models based on different machine learning methods were applied to a fusion dataset containing 24,082 vehicle-level samples. Prediction results indicate that the pre-crash indicators (PCIs) are more influential than the commonly used basic crash information because the average accuracy of six models can be improved by 14.35% after utilizing PCIs. Furthermore, the features’ importance and their marginal effects are interpreted based on parameter estimation, Shapley additive explanation value, and partial dependence. The ΔV, EES, and CMI are identified as the determinant indicators of the potential IR, and their partial distributions are significantly influenced by the crash type and impact position. Based on partial dependence probabilities, the study establishes decision thresholds for PCIs for each severity category for different impact positions, which can serve as a useful reference for developing targeted safety strategies. These results suggest that the proposed method can effectively improve pre-crash IR assessment, which can be readily transferred to safety-related modeling in an active traffic management system.

Prediction based activation of vehicle safety systems – a contribution to improve to occupant safety by validation of pre-crash information and crash severity plus restraint strategy prediction

Abstract This work presents an approach to activate vehicle safety systems based on a pre-crash prediction. It focuses on the three major topics namely, crash validation, the geometry-estimation, and the crash severity and restraint strategy prediction. The topic of crash validation is administered by a proposal of a novel contact-based validation sensor along with the experimental investigations for comparison with other sensors.

Construction of Collision-Type Prediction Models Based on Pre-crash Data for Advanced Driver Assistance Systems

ABSTRACT: Collision-type prediction models based on pre-crash information are important because there is a relationship between collision type, avoidance operations, and occupant injuries. Thus, they can be applied to autonomous driving systems (ADS) or advanced driver assistance systems (ADAS) to prevent serious accidents or minimize damage during collisions. In this study, we investigated the application of collision-type prediction models based on several machine learning methods and compared their performance to determine the best model based on their f1 scores. The results revealed that the light gradient boosting machine (LGBM) model had a high f1 score that exceeded 0.92, which implied that it could potentially be used for ADS and ADAS applications. Furthermore, a brief analysis was performed on the ranking of various factors, which provided useful insight into the significance of several pre-crash factors and their distributions.

FEM-based Methodology for Crash Severity Estimation in Frontal Crash Scenarios

With the technological development of forward-looking sensors, researchers are exploring their use not only for advanced driver assistance systems but also to gain important pre-crash information. Based on this pre-crash information, if the occupant motion inside the vehicle structure can be predicted for the oncoming crash scenario, then an optimal restraint strategy can be planned before the crash. This paper introduces a two-step FEM simulation based methodology for predicting the occupant severity in head-on crash scenarios. In the first step, we simulate the vehicle level model with different impact positions and relative approach angles. The results of these simulations, linear velocities in the longitudinal and lateral direction and angular velocities (roll, pitch, and yaw) during in-crash phase are the loading conditions for next simulation step (occupant level). This step simulates the motion of the driver in different crash scenarios. In this paper, we investigate the head, neck, and chest injury risks from vehicle-to-vehicle crash both traveling at 50 kilometers per hour. Prediction of the head injury criterion, identifying the cases where additional deployment of side-airbags and discussion of injury criteria with contour plots are the main outcome of this paper.

Second strategic highway research program naturalistic driving study methods

Collecting detailed, objective pre-crash information has been a challenge for the transportation community. While simulator studies, crash databases, and police accident reports/investigations help provide data that contribute to our overall understanding of crash risk, such studies are limited in that they cannot provide accurate information about the seconds and factors immediately preceding a crash or near-crash as such events occur in the real world. The naturalistic driving study method helps to fill this knowledge gap, with participants’ personal vehicles equipped with discreet instrumentation that includes cameras and sensors that continually collect kinematic, environmental, and behavioral data as drivers go about their normal driving routines. The naturalistic driving study of the Second Strategic Highway Research Program (SHRP 2) constitutes the largest such study with video to date. It captured more than 1900 light-vehicle crashes and 6900 near-crashes, making it a robust resource for researchers who wish to find solutions that help mitigate the occurrence of safety-critical events on national highways. The study database comprises more than 32 million miles of continuous data from over 3500 drivers across six data collection sites in the United States, allowing also for the study of traffic operations and management. The methods used to design this large-scale study, recruit participants, and highlights of the collected data are included; overall management of the study and its strengths and weakness are also discussed.

What can the drivers’ own description from combined sources provide in an analysis of driver distraction and low vigilance in accident situations?

Accident data play an important role in vehicle safety development. Accident data sources are generally limited in terms of how much information is provided on driver states and behaviour prior to an accident. However, the precise limitations vary between databases, due to differences in analysis focus and data collection procedures between organisations. If information about a specific accident can be retrieved from more than one data source it should be possible to combine the available information sets to facilitate data from one source to compensate for limitations in the other(s). To investigate the viability of such compensation, this study identified a set of accidents recorded in two different data sources. The first data source investigated was an accident mail survey and the second data source insurance claims documents consisting predominantly of insurance claims completed by the involved road users. An analysis of survey variables was compared to a case analysis including word data derived from the same survey and filed insurance claims documents. For each accident, the added value of having access to more than one source of information was assessed. To limit the scope of this study, three particular topics were investigated: available information on low vigilance (e.g., being drowsy, ill); secondary task distraction (e.g., talking with passengers, mobile phone use); and distraction related to the driving task (e.g., looking for approaching vehicles). Results suggest that for low vigilance and secondary task distraction, a combination of the mail survey and insurance claims documents provide more reliable and detailed pre-crash information than survey variables alone. However, driving related distraction appears to be more difficult to capture. In order to gain a better understanding of the above issues and how frequently they occur in accidents, the data sources and analysis methods suggested here may be combined with other investigation methods such as in-depth accident investigations and pre-crash data recordings.

A Comparative Study between GMLAN Speed and GPS Reported Vehicle Speed by Vehicle Maneuver

Some GM (General Motors) vehicles are using a GMLAN (General Motors Local Area Network) communication protocol for control and diagnostics. The airbag control module uses vehicle speed information from the GMLAN to record the vehicle speed as pre-crash information. In order to use the vehicle speed information for crash reconstruction purposes, it helps to be able to understand the accuracy of the data. The actual vehicle speed is not expected to be the same as the GMLAN indicated speed in some situations like a spin or if there is hard braking. This paper compares the actual vehicle speed and vehicle speed information during specific vehicle maneuvers. Actual vehicle speed is calculated from a GPS sensor, while GMLAN vehicle speed is calculated from transmission output sensor by the Engine control module (ECM). Vehicle maneuvers defined as Mode #1, Mode #2, Mode #3. The Mode #1 maneuver simulates wheel lock-up and skidding during Hard-braking at a specific speed. The Mode #2 maneuver simulates a 90 degree turn using a J-turn maneuver at a specific speed. The Mode #3 maneuver simulates a 180 degree turn using a spin type of maneuver at a specific speed. The study then compares the GMLAN speed and GPS speed to see what speed difference exists between them. The results of this paper are applicable to GM vehicles only. This paper catalogs the performance and limitations of two vehicles as useful reference for crash reconstructions where there is a need to understand the speed indicated in the pre-crash section of the SDM data.

Design of an Airbag Deployment Algorithm Based on Precrash Information

Airbag systems have become an essential safety device for guaranteeing the physical well-being of drivers and their passengers. Unlike other safety devices, airbags are used as the last resort in a collision, and because they are directly linked to the life of the driver and passengers, the proper functioning of the system is an issue of paramount importance. Hence, to ensure the precision and reliability of airbag operation, it is necessary to design a robust crash algorithm. Currently, several companies are working to achieve an optimal deployment time for airbags in collisions by diversifying the type and location of crash-related sensors. Nevertheless, several problems must still be confronted. For instance, when a vehicle operates off road or when the sensor inside the airbag control unit (ACU) receives a powerful shock, the vehicle's airbags may inadvertently deploy, although no collision has occurred, because a crashlike signal is delivered to the ACU. Alternately, in a collision situation that requires airbag deployment, the crash algorithm may make an erroneous judgment with regard to the collision configuration and miss the time frame for airbag deployment or fail to deploy the airbags altogether. Such problems can be attributed to the following two major causes: 1) Only signals produced through crash tests are used in the design of crash algorithms, and 2) the algorithms themselves only utilize postcrash input from the relevant sensors. To resolve these issues, this paper proposes a precrash algorithm that generates information about the crash scenarios before a collision has occurred. The purpose of the precrash algorithm is to make judgments about the impending collision configuration prior to impact by estimating the behavior of frontal objects and to communicate this information to the crash algorithm to enable correct recognition of the crash scenarios. This paper also proposes a crash algorithm based on crash-related sensors for the verification of and interfacing with the proposed precrash algorithm. The limitations of existing crash algorithms, which deploy airbags using only postcrash signals, were resolved through the development of an integrative crash algorithm that combines the precrash and crash algorithms to reflect precrash information. The developed algorithm was verified by running a wide range of simulations using CarSim, based on data from real crash tests. The results showed that, compared with independently using the crash algorithm, adding precrash information estimation significantly improved the reliability of airbag deployment.

Design of an Airbag Deployment Algorithm Using a Radar Sensor

An airbag system has been a fundamental safety equipment for saving lives of a driver and passengers. This system has some problems even though it is very efficient for protecting drivers or passengers. After a crash is occurred, airbags must be deployed in a proper time along the crash situation. But it is tricky to deploy at the exact time depending on the situations. These tricky situations happen especially for the airbag deployment algorithm using only the post-crash signals. In general, the airbag deployment algorithm uses airbag control unit(ACU) X-Y sensor and frontal impact sensor(FIS). And it discriminates crash situations along these sensors signals. However, during the crash process, these sensors are broken, rotated or moved. That causes the malfunction of airbag deployment algorithm. In this study, a new airbag deployment algorithm is developed which is enhanced with a radar sensor signal. Using pre-crash information from the radar sensor, the algorithm can judge a crash situation before a real crash is occurred and can revise the post-crash signal.