Precise And Predictable Automated Driving Cognitive Approach Based On Complex Conditions

In recent years,driven by the new round of technology revolution represented by Artificial Intelligence,the automobile industry is growing rapidly.Autonomous vehicle provides new solutions to problems such as traffic safety,traffic congestion,fuel consumption and air pollution.In automated driving system framework,environmental cognition is the crucial prerequisite and step to ensure the smooth and orderly work of automated driving system.Real-world road condition must be taken into consideration in order to achieve reliable environmental cognition analysis.However,current research based on simple road condition and single sensor could not meet the requirement of multitarget object recognition and tracking tasks,and also failed to understand the behavior purpose of the target object.To solve the aforementioned problems,the thesis establishes a precise and predictable automated driving environment cognition framework based on complex operating conditions,and explores the key technologies involved in the process of environmental cognition.To deal with the potential danger in various and complex road conditions such as arch bridge,ramp,speed bump,potholes etc.,under which background point cloud may be perceived as foreground point cloud mistakenly,or foreground obstacle could be easily recognized as background,the thesis proposes a background point cloud segmentation method based on three-dimensional grid analysis,connected components labeling method based on pixel growth and object classification method based on contour and trajectory.It can realize efficient and precise obstacle identification in dynamic road environment,and it is validated by the test data of actual road scenes.In order to arrange and use the detected information reasonably and obtain consistency in cognition of detected target,a multi-target tracking method based on DSEKF theory is proposed.The method combines the redundancy or complementary information of time and space using corresponding criterion based on the DS evidence theory and Kalman filter theory,and designs the algorithm for target life cycle management,target correlation matching and target tracking filtering.The test results show that the algorithm is able to realize the multi-target object tracking with high precision.In order to ensure the rationality of behavior prediction,to simplify traditional forecasting model and shorten the development period of the behavior prediction system,a multiple-type target prediction method which integrates road model with behavior intention.By building the traffic model of actual road condition,and recognizing the intension of target object in the environment,the algorithm is able to plan a road network in advance which consists of the map and topological structure of the road.What all the autonomous vehicles need to do is to identify the most likely one or several paths from the historical trajectory of target object.Test results prove that the method is efficient since it successfully avoids the complex analysis of target object’s behavior intention and driving path.In order to verify the validity of the precise and predictable automated driving environment cognitive approach under complicated operating conditions,a sensor algorithm validation dataset of the automated driving system,as well as a completed road scene database is built,which covers conditions including urban area,highway and loop scenes.The target recognition,target tracking and behavior prediction algorithm are tested using the dataset.The experimental results show that the precise and predictable automated driving environment cognitive approach presented in the thesis is effective for target recognition and target tracking in a variety of complex environment conditions.Moreover,the approach is able to make prediction for motor vehicle behavior,cyclist behavior and pedestrian behavior,which provides more accurate information for the decision making step of automated driving system.