On Development of Intellidrive-based Red Light Running Collision Avoidance System

Intellidrive offers great potential for various new safety and mobility applications. This paper reports on an Intellidrive application for signalized intersection safety, where signals can dynamically adapt to hazardous conditions in order to avoid Red-Light-Running (RLR) related collision. As an important portion of the intersection collision avoidance system, a model for the RLR prediction based on Intellidrive vehicle-toinfrastructure communications is proposed. Both the isolated vehicle motion and car-following status are used for the road-side unit to predict signal violation. The prediction is an improvement to a previous RLR detection means reported to-date. Issues related to such a dynamic system with integration of infrastructure (signal controller) and Intellidrive communications are discussed. The system performance is then emulated using data collected from infrastructure-based sensor as an alternative source to Intellidrive data which is not widely available yet. Moreover, from a field experiment, we show the accurate and timely RLR prediction potential that will be provided by Intellidrive-equipped vehicles. INTRODUCTION Traffic crashes are the most significant cause of preventable death and injury in North America. The National Highway Traffic Safety Administration (NHTSA) reports that in 2005 alone, nearly 9,200 people were killed and approximately one million people were injured in intersection-related crashes. These intersection-related crashes account for about 40%-45% of all crashes. According to 2005 data from NHTSA’s Fatality Analysis Reporting System, crashes caused by red light running (RLR) resulted in an estimated 805 fatalities (1 ). The national initiative of Vehicle Infrastructure Integration (VII; recently re-named as Intellidrive) reveals great potential in intersection safety applications (2 , 3 ). Important national projects include the Cooperative Intersection Collision Avoidance System (CICAS)-Violations (CICAS-V) , developed by various manufacturers, CICAS-Stop Sign Assist (CICAS-SSA), conducted by Minnesota DOT and University of Minnesota, CICAS-Signalized Left-Turn Assist (CICAS-SLTA) and Traffic Signal Adaptation (CICAS TSA), being developed by a partnership between Caltrans and California PATH at the University of California, Berkeley (4 , 5 ). The research in this paper is supported by CICAS-SLTA and TSA, conducted at California PATH. The CICAS TSA system intends to dynamically adapt the traffic signal in the form of extending an all-red period by a few seconds to avoid collisions due to signal violations. A dynamic TSA system comprises several elements, including Intellidrive-based state-map building, signal violation (hazardous situation) prediction and countermeasures. While these elements are all critical and interrelated, the RLR prediction is the most critical challenging element, as reliable detection of RLR behavior is the prerequisite for successfully executing safety countermeasures. Lanjun Wang: On Development of Intellidrive-based RLR Collision Avoidance System 2 Researchers have proposed various models for vehicle motion prediction at signalized intersections (6 , 7 , 8 ). Sheffi and Manhamassani (6 ) assumed that a "critical time" is normally distributed among drivers, and developed a statistical model of driver behavior at a high speed isolated intersection after the onset of the yellow light. California PATH recently proposed an extended probabilistic model to address vehicle speed and acceleration, and to address two different kinds of errors when optimizing system performance (8 ). This paper focuses on an Intellidrive-based violation prediction algorithm which is specially designed for all-red extension for Intellidrive-enabled vehicles in order to improve RLR prediction performance for the CICAS TSA system. This work is a continuation of the algorithm reported in (8 ) in the sense that an elaborate model which takes into account both the isolated vehicle motion as well as the car-following status to form a prediction algorithm which works for Intellidrive-enabled vehicles. The study introduces another dimension of information, which is the car-following status measured by the speed difference of the leading vehicle and the following vehicle at a certain time (relative to red-onset). From the field observations (9 ), it is found that over 60% of the vehicles were in a platoon (in the sense that headways were less than 3 seconds). It is important to note that the driver’s decision making procedure is not only impacted by the signal phase, but by the leading vehicle as well. This additional piece of information helps to better understand driver reaction to the signal phase change, especially for the cases when drivers do not decelerate very hard after the onset of the yellow light and thus whose stop-go decisions are not easily predicted. This paper is organized as follows. Section 2 presents a brief description of the system, contains a conceptual diagram of the whole CICAS TSA system and how the RLR prediction algorithm works, and especially discusses the performance of Intellidrive. Section 3 verifies the improved prediction method with data from an infrastructure-based sensor, and proves the Intellidrive’s advantages. Finally, section 4 offers conclusions to this paper. INTELLIDRIVE BASED RLR PREDICTION FOR CICAS TSA When an Intellidrive-equipped vehicle is approaching a signalized intersection, the on-board equipment (OBE) sends vehicle information to the road-side equipment (RSE). Typical vehicle data includes vehicle trajectory data (speed, position, time stamp, etc). Together with such data, the RSE also has information about the infrastructure, such as geometrical intersection description (GID), and signal phase and timing (SPAT). A processor in the RSE will determine whether a violation will occur and request an all-red extension for impending violations. The conceptual structure of the Intellidrive-based TSA system is shown in Fig. 1. For each approaching vehicle i , we have a series of speeds: {vi (1) , vi (2) , ..., vi (K)} Lanjun Wang: On Development of Intellidrive-based RLR Collision Avoidance System 3 Signal Controller Vehicle OBE