Rini Sherony

Evolving a real-world vehicle warning system

Many serious automobile accidents could be avoided if drivers were warned of impending crashes before they occur. Creating such warning systems by hand, however, is a difficult and time-consuming task. This paper describes three advances toward evolving neural networks with NEAT (NeuroEvolution of Augmenting Topologies) to warn about such crashes in real-world environments. First, NEAT was evaluated in a complex, dynamic simulation with other cars, where it outperformed three hand-coded strawman warning policies and generated warning levels comparable with those of an open-road warning system. Second, warning networks were trained using raw pixel data from a simulated camera. Surprisingly, NEAT was able to generate warning networks that performed similarly to those trained with higher-level input and still outperformed the baseline hand-coded warning policies. Third, the NEAT approach was evaluated in the real world using a robotic vehicle testbed. Despite noisy and ambiguous sensor data, NEAT successfully evolved warning networks using both laser rangefinders and visual sensors. The results in this paper set the stage for developing warning networks for real-world traffic, which may someday save lives in real vehicles.

Neuroevolution of an automobile crash warning system

Many serious automobile accidents could be avoided if drivers were warned of impending crashes before they occurred. In this paper, a vehicle warning system is evolved to predict such crashes in the RARS driving simulator. The NeuroEvolution of Augmenting Topologies (NEAT) method is first used to evolve a neural network driver that can autonomously navigate a track without crashing. The network is subsequently impaired, resulting in a driver that occasionally makes mistakes and crashes. Using this impaired driver, a crash predictor is evolved that can predict how far in the future a crash is going to occur, information that can be used to generate an appropriate warning level. The main result is that NEAT can successfully evolve a warning system that takes into account the recent history of inputs and outputs, and therefore makes few errors. Experiments were also run to compare training offline from previously collected data with training online in the simulator. While both methods result in successful warning systems, offline training is both faster and more accurate. Thus, the results in this paper set the stage for developing crash predictors that are both accurate and able to adapt online, which may someday save lives in real vehicles.

An Extreme Learning Machine-based pedestrian detection method

Pedestrian detection is a challenging task due to the high variance of pedestrians and fast changing background, especially for a single in-car camera system. Traditional HOG+SVM methods have two challenges: (1) false positives and (2) processing speed. In this paper, a new pedestrian detection method using multimodal HOG for pedestrian feature extraction and kernel based Extreme Learning Machine (ELM) for classification is presented. The experimental results using our naturalistic driving dataset show that the proposed method outperforms the traditional HOG+SVM method in both recognition accuracy and processing speed.

Estimation of the vehicle-pedestrian encounter/conflict risk on the road based on TASI 110-car naturalistic driving data collection

Modeling vehicle-pedestrian interactions in the road environment is essential to develop pedestrian detection and pedestrian crash avoidance systems. In this paper, one novel approach is proposed to estimate the vehicle-pedestrian encountering risk in the road environment based on a large scale naturalistic driving data collection. Considering the difficulty to record actual pedestrian crashes in the naturalistic data collection, the encountering risk is estimated by the chances for driver to meet with pedestrian in the roadway as well as the chances for the driver and pedestrian to get into a potential conflict. Effects of different scenarios consisting of road conditions, pedestrian behaviors, and pedestrian numbers on the risk levels are also evaluated, and significant results are provided.

Automatic categorization-based multi-stage pedestrian detection

Pedestrian safety has become an important issue for automobile design. Although a lot of research has been done or is ongoing in developing in-car camera-based pedestrian protection systems, robust and reliable in-car camera based pedestrian analysis is still very challenging, especially for real-time systems or large scale dataset analysis. In this paper, we propose a new pedestrian detection and analysis system based on automatic categorization. A category-based multi-stage pedestrian detection and data analysis approach is developed to efficiently process the extremely large scale driving data collected in this research. The experimental results on part of the collecting dataset show that the proposed method is promising.

Mannequin Development for Pedestrian Pre-Collision System Evaluation

Pedestrian PCS (pre-Collision System), as a novel active pedestrian safety equipment to reduce pedestrian fatalities, was introduced by several vehicle manufactures in recent years. How to evaluate this system is still in study. This paper describes the development of a set of mannequins for pedestrian PCS evaluation. These mannequins represent relevant physical properties of human beings that are used by most common PCS sensors in vehicles for pedestrian detection. Three different mannequin sizes were generated to represent child, fit adult and obese adult of U.S. pedestrians. These sizes were generated using a K-means clustering method with crash weighted NHANES (National Health and Nutrition Examination Survey) data. To ensure that the mannequin provides the same radar cross section (RCS) as a pedestrian in the view of a 77GHz automotive radar, a special mannequin skin was developed. To make the mannequin move like a real human in the view of PCS cameras, mannequins were configured with 6 degrees of freedom articulation, 2 for shoulders, 2 for hips, and 2 for knees. The mannequin limbs are detachable at crash in order to protect the mannequin motion driving system. To ensure that the mannequins move like a real human, a gait planning method was adopted to generate different walking speed based on the gaiting research in medical fields. Other mannequin features are to support various test setups, low weight for test vehicle safety, and crash forgivingness. The mannequins were used in hundreds of PCS test runs at different crash angles and vehicle speeds.

Potential Occupant Injury Reduction in the U.S. Vehicle Fleet for Lane Departure Warning–Equipped Vehicles in Single-Vehicle Crashes

Objective: Single-vehicle collisions involve only 10 percent of all occupants in crashes in the United States, yet these same crashes account for 31 percent of all fatalities. Along with other vehicle safety advancements, lane departure warning (LDW) systems are being introduced to mitigate the harmful effects of single-vehicle collisions. The objective of this study is to quantify the number of crashes and seriously injured drivers that could have been prevented in the United States in 2012 had all vehicles been equipped with LDW. Methods: In order to estimate the potential injury reduction benefits of LDW in the vehicle fleet, a comprehensive crash and injury simulation model was developed. The models basis was 481 single-vehicle collisions extracted from the NASS-CDS for year 2012. Each crash was simulated in 2 conditions: (1) as it occurred and (2) as if the driver had an LDW system. By comparing the simulated vehicles off-road trajectory before and after LDW, the reduction in the probability of a crash was determined. The probability of a seriously injured occupant (Maximum Abbreviated Injury Score [MAIS] 3+) given a crash was computed using injury risk curves with departure velocity and seat belt use as predictors. Each crash was simulated between 18 and 216 times to account for variable driver reaction, road, and vehicle conditions. Finally, the probability of a crash and seriously injured driver was summed over all simulations to determine the benefit of LDW. Results and Conclusions: A majority of roads where departure crashes occurred had 2 lanes and were undivided. As a result, 58 percent of crashes had no shoulder. LDW will not be as effective on roads with no shoulder as on roads with large shoulders. LDW could potentially prevent 28.9 percent of all road departure crashes caused by the driver drifting out of his or her lane, resulting in a 24.3 percent reduction in the number of seriously injured drivers. The results of this study show that LDW, if widely adopted, could significantly mitigate a harmful crash type. Larger shoulder width and the presence of lane markings, determined by manual examination of scene photographs, increased the effectiveness of LDW. This result suggests that highway systems should be modified to maximize LDW effectiveness by expanding shoulders and regularly painting lane lines.

Pilot study on pedestrian step frequency in naturalistic driving environment

Investigation of pedestrian step frequency is essential for analyzing walking gaits and pedestrian behaviors. However, most research about step frequency is performed in labs or manually controlled experimental environment, which greatly limits the utilization of the results to analyze and/or predict real pedestrian behaviors. This study investigates the step frequencies of pedestrian in naturalistic driving environment. The mean step frequency values and distribution are studied in all cases and separately for road crossing cases only. Furthermore, comparisons of pedestrian step frequencies are made considering three different impact factors. The results have shown that in real world, people tend to use higher step frequencies when crossing the road, especially when the vehicle is moving towards the pedestrian or when the pedestrians are crossing without right-of-way.

77 GHz radar scattering properties of pedestrians

Using radars for detecting and tracking pedestrians has important safety and security applications. Most existing human detection radar systems operate in UHF, X-band, and 24GHz bands. The newly allocated 76-77 GHz frequency band for automobile collision avoidance radar systems has once again raised great interest in pedestrian detection and tracking at these frequencies due to its longer detection range and better location accuracy. The electromagnetic scattering properties must be thoroughly understood and analyzed so a catalog of human scattering can be utilized for intensive automotive radar testing. Measurements of real human subjects at these frequencies is not a trivial task. This paper presents validation between the measured radar cross section (RCS) patterns of various human subjects and a full-wave EM simulation. The RCS of a human is shown to depend on a number of factors, such as posture, body shape, clothing type, etc.

Potential of intersection driver assistance systems to mitigate straight crossing path crashes using U.S. nationally representative crash data

Intersection Advanced Driver Assistance Systems (I-ADAS) are active safety systems that have the potential to help prevent/mitigate crashes and injuries in intersection crashes. I-ADAS may use side-looking sensors, e.g. radar and lidar, in order to detect potential collisions with vehicles from crossing paths. The success of I-ADAS depends on the range and azimuth capabilities of these sensors. In order to specify the capabilities of sensors for an I-ADAS, designers need a distribution of range and azimuth between vehicles as they enter intersections prior to crashes. This study generated range and azimuth distributions using crash data from the National Motor Vehicle Crash Causation Survey (NMVCCS) for vehicles just prior to entering the intersection in straight crossing paths (SCP) crashes. Using the reconstructions and specifications in existing radar technology, the potential crash mitigation benefits of this technology were determined. Three radar-based I-ADAS were analyzed using published sensor specifications. The sensors included a wide beam, intermediate beam, and narrow beam. The wide beam I-ADAS was found to detect 20.3% of oncoming vehicles, the intermediate beam was found to detect 89.2% of oncoming vehicles, and the narrow beam was found to detect 98.3% of oncoming vehicles. The results indicate that a narrow beam I-ADAS is the most capable because of its long range detection ability. These results have practical relevance for the design and implementation of I-ADAS.