Environment Perception for Autonomous Driving : A 1/10 Scale Implementation Of Low Level Sensor Fusion Using Occupancy Grid Mapping

Sammanfattning: Autonomous Driving has recently gained a lot of recognition and provides challenging research with an aim to make transportation safer, more convenient and efficient. This emerging technology also has widespread applications and implications beyond all current expectations in other fields of robotics. Environment perception is one of the big challenges for autonomous robots. Though a lot of methods have been developed to utilize single sensor based approaches, since different sensor types have different operational characteristics and failure modes, they compliment each other. Different sensors provide different sets of data, which creates difficulties combining information to form a unified picture. The proposed solution consists of low level sensor fusion of LIDAR and stereo camera data using an occupancy grid framework. Bayesian inference theory is utilized and a real time system has been implemented on a 1/10 scale robot vehicle. The result of the thesis shows that it is possible to use a 2D LIDAR and stereo camera to build a map of the environment. The implementation focuses on the practical issues like blind spots of individ sensors. Overall, the fused occupancy grid gives better result than occupancy grids from individual sensors. Sensor confidence is higher for the camera since frequency of mapping of a 2D LIDAR is low