A Combined Approach for Object Recognition and Localisation for an Autonomous Racecar

Abstract: With autonomous vehicles being a hot topic for research it has also become an interest in theworld of motor sport. To be able to run a vehicle autonomously it needs to know what the currentpose of the vehicle is and what the environment looks like. This thesis aims to solve this problemusing SLAM and object detection with 2D LiDAR and camera as sensor input, looking at theperformance in terms of accuracy and latency.The object detection problem was repurposed as an object recognition problem by utilising the2D LiDAR for cone candidate extraction which was projected onto the camera image andverified by a Convolutional Neural Network (CNN). Two different CNN architecture were used,MobileNet and a minimalistic architecture with less than 7 layers. The best performing CNNwith four convolutional layers and two fully connected layers reached a total of 87.3% accuracywith a classification time of 4.6ms on the demonstrator constructed.Three different SLAM algorithms were implemented, Pose Graph Optimization, Rao-Blackwellized Particle Filter and Extended Kalman Filter (EKF). When tested on thedemonstrator the EKF solution showed the best results with a mere 20mm average error invehicle position and 39mm average error in cone position. Further, the end-to-end timing of theEKF algorithm was the fastest at an average of 32ms.The two best performing algorithms were combined for an evaluation, with the output of theCNN as input to the EKF. The performance was measured to an average error of 19mm for theposition and 51mm for the cones. Further, the latency was only increased by the 4.6ms that theCNN required for classification, to a total of 36.54ms.