Visual Odometry for Vehicles’ Undercarriage 3D Modelling
This work describes a part of a project developing a vehicles’ undercarriage security scanner based only on cameras. The scanner is used to a security check of a vehicle’s undercarriage and is typically installed at an entrance to a strategic compund. The security scanner reconstruct a 3D model of a vehicle’s undercarriage from a sequence of a multi-camera stereo images. To get a complete model we need to stitch particular parts of the 3D model via transformations between particular vehicle positions in which images are captured. The method for getting these transformations is presented in this paper. The task of computing trajectory from a sequence of camera images is called visual odometry (VO). Usually, the camera is placed on a moving object and tracks its position. In our case, the camera is fixed and viewing a moving vehicle, but the task is the same. In the first part, there is a comparison of feature detectors and their parameters based on experimental data because the images properties of undercarriage are different from ordinary surroundings used by the most of VO methods. Undercarriages do not contain a lot of features, and there are many low-textured surfaces. In the second part, the proposed VO method is described. It is based on the best feature from the first part, which serves to search corresponding points between images. It uses 3D to 2D VO method to compute the transformation between consecutive frames. In this method, 3D points are triangulated from the previous pair of stereo camera images, and they are reprojected to one of the actual images. The method finds the transformation of 3D positions which minimizes reprojection error. The method was developed with respect to requirement of almost real-time computing time and low-texture environment. Finally, this method was evaluated on realistic data acquired with ground truth position.
KeywordsVisual odometry Stereo vision Security Car inspection 3D modelling
The work described in this paper was funded by Ministry of Interior of the Czech Republic under the project number VI2VS/461. The work of Tomáš Pivoňka was also supported by the Grant Agency of the Czech Technical University in Prague, grant No. SGS18/206/OHK3/3T/37.
- 3.Cvišić, I., Petrović, I.: Stereo odometry based on careful feature selection and tracking. In: 2015 European Conference on Mobile Robots (ECMR), pp. 1–6 (2015). https://doi.org/10.1109/ECMR.2015.7324219
- 4.Geiger, A.: Visual odometry/SLAM evaluation 2012. In: The KITTI Vision Benchmark Suite. http://www.cvlibs.net/datasets/kitti/eval_odometry.php. Accessed 25 July 2018
- 7.Bradski, G.: The OpenCV Library. Dr. Dobb’s Journal of Software Tools (2000)Google Scholar
- 8.Kneip, L., Scaramuzza, D., Siegwart, R.: A novel parametrization of the perspective-three-point problem for a direct computation of absolute camera position and orientation. In: CVPR 2011, pp. 2969–2976 (2011). https://doi.org/10.1109/CVPR.2011.5995464
- 9.Pivoňka, T.: Visual odometry for dynamic image reconstruction. Diploma thesis (2018). https://dspace.cvut.cz/handle/10467/77042