A Scheme and Algorithm for Automotive Automatic Collision Avoidance System
The importance of automobiles in our life is indubitable, but the amount of traffic accidents caused by automotive is more and more also.According to 2002 annual report of U.S. Federal Highway Administration, more than 42,000 Americans died as a result of 6.8 million crashes on their nations roadways in 2001. On average, a person was injured in one of these crashes every 10 seconds, and someone was killed every 12 minutes. As automobiles proliferating, the threat of traffic accidents to the human is attracting focus to research and develop intelligent transportation system (ITS) to save lives, and intelligent vehicles is the core of the ITS. Automotive Automatic Collision Avoidance System (AACAS) proposed in this paper is a significant scheme of intelligent vehicles.One of great causes of humans research on automobile is to make it more intelligent. The west advanced countries led by America high on researching on the Autonomous Land Vehicles (ALVs) to achieve the final aim of unmanned war, The famous DEMO series program is just an important part of American Scout Unmanned Ground Vehicles.Not only to civil application but also to military requirement, the research on intelligent vehicles is attaching more and more attentions. The scheme and algorithms for AACAS proposed in this paper is an important item in intelligent vehicles. Based on the sensor device and the requirements of our system, we proposed and designed the AACAS principle framework and the sensor subsystem and designed software functional modules to implement it. The software functional module is the core of our system, and we developed a set of algorithms, such as data fusion, target selection, threat assessment and collision avoidance, to implement our scheme.On our experimental vehicle, we constructed an environment sensing system which was composed of external environment sensing subsystem and the vehicle localization subsystem, the former consists of a CCD camera and a laser range finder (LRF), the later is the GPS/INS combined localization equipment. The CCD camera and LRF are used to construct a 2D and 3D hybrid vision system,the CCD camera installed on the top of the vehicle was used to detect lines in the environment, such as road lane edges etc, the laser range finder installed in the front of the vehicle body about 30cm high was used to detect the obstacles in front of the vehicle, such as pedestrian and cars. The GPS/INS combined localization device was used to sense the vehicles position and orientation anytime.The AACAS principal framework proposed in this paper was conformed to the 4-D/RCS standard architecture for Unmanned Ground Vehicles. It consists of three parts, including sensing part, world modeling part and vehicle control part. In the sensing part, the GPS/INS can sense the vehicles position and orientation in the world anytime, and the CCD camera was mainly used to track lane edges, the LRF was mainly used to detect obstacles. In the vehicle control part, control algorithms were used to respond to the action of world modeling to the vehicle body such as turning, deceleration etc. In this paper, we mainly proposed a world modeling scheme consists of data fusion, target selection,threat assessment and collision avoidance strategy to implement our system,which is the core of AACAS.The basic idea of our world modeling part is to integrate temporal and spatial information collected by sensors to obtain a uniform and reliable environment map and give a planning to the vehicles motion. After temporally fusing, lane edges and obstacles wobble caused by sensors inherent error and environment noise can be eliminated to the least. Through the spatial fusion of lane edges and obstacles, the obstacles which threat vehicle can be selected. As to every obstacle, its threat to the vehicle body can be evaluated based on the threat assessment function which is defined as a function related to obstacles size and position relative to the vehicle and the vehicles status, such as vehicles position,speed and acceleration etc.Based on the threat assessment of every obstacle to the vehicle, we select those threat vehicle most to steer around it or stop the vehicle. When the threat of obstacles to the vehicle body dose exist, AACAS must give an action to avoid collision. According to the threat assessment, two strategies can be adopted, that is pure stopping maneuver and steer around turn maneuver, and their time of reaction, stop distance and turning radius etc can be computed in our collision avoidance functional module. Based on our AACAS principal framework and corresponding algorithms,simulation experiments and physical experiments are conducted in the proving ground. Our sensing subsystem can successfully detect lanes and manmade obstacles on the road and the world modeling system can properly integrate multisensor information to select those that threat vehicle most and steer around them. Experiment results show that the proposed scheme and algorithms are effective and reliable.In the United States, Japan and some European countries, they apply themselves to promoting automotive intelligence.In the future, we will research the vehicles mobility to make it steer around moving obstacles such as pedestrian and cars etc also.
Automotive Automatic Collision Avoidance System sensor fusion threat assessment collision avoidance
Huajun Liu
Computer Science Dept., Nanjing University of Sci.& Tech., Nanjing, China, 210094
国际会议
首届嵌入式软件与系统国际会议(Proceedings of the First International Conference on Embedded Software and System)
杭州
英文
557-558
2004-12-09（万方平台首次上网日期，不代表论文的发表时间）