Search for: All records

This work focuses on finding the extrinsic parameters (rotation and translation) between the lidar and the stereo camera setups. We use a planar checkerboard and place it inside the Field-of-View (FOV) of both the sensors, where we extract the 3D plane information of the checkerboard acquired from the sensor’s data. The planes extracted from the sensor’s data are used as reference data sets to find the relative transformation between the two sensors. We use our proposed method Correntropy Similarity Matrix Iterative Closest Point (CoSM-ICP) algorithm to estimate the relative transformation. In this work, we use a single frame of the point cloud data acquired from the lidar sensor and a single frame from the calibrated Stereo camera point cloud to perform this operation. We evaluate our approach on a simulated dataset since it has the freedom to evaluate under multiple configurations. Through results, we verify our approach under various configurations. 

This work attempts to answer two problems. (1) Can we use the odometry information from two different Simultaneous Localization And Mapping (SLAM) algorithms to get a better estimate of the odometry? and (2) What if one of the SLAM algorithms gets affected by shot noise or by attack vectors, and can we resolve this situation? To answer the first question we focus on fusing odometries from Lidar-based SLAM and Visualbased SLAM using the Extended Kalman Filter (EKF) algorithm. The second question is answered by introducing the Maximum Correntropy Criterion - Extended Kalman Filter (MCC-EKF), which assists in removing/minimizing shot noise or attack vectors injected into the system. We manually simulate the shot noise and see how our system responds to the noise vectors. We also evaluate our approach on KITTI dataset for self-driving cars.