Localisation and Mapping of Self-driving Vehicles based on Fuzzy K-means Clustering: A Non-semantic Approach

Localisation and mapping are crucial for autonomous vehicles, as they inform the vehicle of where exactly they are in their environment as well as relevant infrastructures within the identified environment. This paper demonstrates the ability of non-semantic features to represent point clouds and use them to explain the environment. Our proposed architecture uses the Fuzzy K-means approach to extract features from LiDAR scenes in order to reduce the feature map and guarantee that the features are identifiable in each frame. Secondly, global mapping is done with the Gaussian Mixture Model (GMM) to facilitate data association between the frames to be mapped and helps localisation tasks to be performed accurately by the particle filter. The performance of the proposed technique is compared to other state of the art methods over different sequences of the Kitti raw dataset with different environmental structures, weather conditions and seasonal changes. The results obtained demonstrates the superiority of the proposed approach in terms of speed and representativeness of features needed for real-time localisation. Moreso, we achieved competitive accuracies compared to other state-of-the-art methods.