Sensor Fusion-Based Target Prediction System for Virtual Testing of Automated Driving System

The perception system is one of the important components of autonomous vehicles, as it provides the information that is required by vehicle control to make decisions on the manoeuvre of the vehicle. The study focuses on the development of target prediction using sensor fusion algorithm for Level 3 autonomous vehicle in Malaysian environment. The sensor fusion algorithm was developed to unify the data from the sensors and obtain useful information, where the closest object around the ego vehicle was determined in the project. In order to display the closest object around the ego vehicle, the relative distances of the objects were calculated. The closest object among the cameras, the closest object in each camera and warning for nearby object were displayed on the output images. To study the performance of sensor fusion algorithm developed in Malaysian traffic, the virtual environment model of MyAV Route A was developed by using RoadRunner. There were two cases developed to observe how would the algorithm perform. The first test case was on target prediction using sensor fusion algorithm on flat road, while the second test case was on target prediction along a slope. It was shown that the algorithm performed well from the two test cases, as the vehicles and pedestrians were detected and displayed successfully with confidence score of above 0.72 and 0.87, respectively, even with views from different angles and locations of cameras.