Sensor fusion with occupancy grid map for obstacle segmentation in intelligent vehicles

Recent advances in UGV technology has surged a great amount of interest and accelerated research activities in this area. One of the main research topics in UGVs is the use of intelligent systems including sensors such as lasers, cameras, GPS and radar systems, which are already being increasingly employed in automotive and defence industry. In this paper, we summarize our research on the development of mapping algorithms for the "Unmanned Ground Vehicle (UGV) Project", carried out in ITU Mechatronics Education and Research Center. Testing a system in real time with all sensors working correctly is both difficult and expensive. Therefore to test our mapping algorithms, we used Webots, which is a professional simulation program by Cyberbotics. After constructing an occupancy grid map, we accepted the map as an image and applied image processing techniques. We used median filter to decrease the noise, dilation is applied to reduce sensor and measurement faults. Then, obstacles are segmented and enumerated by region growing algorithm. Finally, in order to decrease the computational load, each obstacle is described by only its mean and boundary values for further use in the path planning algorithm.