Computer Vision
Camera models, features, geometry — robots that see.
Why it matters in robotics
Most robots perceive the world through cameras, so interviewers probe whether you understand how pixels map to 3D geometry — camera intrinsics/extrinsics, projection, and triangulation are the bridge from images to a metric world model. Perception and SLAM roles routinely ask you to reason about epipolar geometry, feature matching, RANSAC for outlier rejection, and pose estimation, since these underpin visual odometry and mapping. Strong candidates connect the classical geometry (the "why") to practical pipelines (OpenCV calibration, SfM, visual SLAM) rather than treating vision as a black-box deep net.
Application focus
The same topic, tailored to the robot you're building. Your choice is remembered across the roadmap and every topic.
At a glance
The classical computer-vision pipeline: raw pixels become sparse features, feature correspondences constrain multi-view geometry, and geometry yields camera pose plus 3D structure.
What to study
- ✓Camera models: pinhole/perspective projection, intrinsic & extrinsic matrices, lens distortion, and calibration
- ✓Features and matching: detectors/descriptors (SIFT, ORB), correspondence, and RANSAC for robust outlier rejection
- ✓Two-view geometry: epipolar constraint, fundamental & essential matrices, pose recovery, and triangulation
- ✓Structure from Motion & visual SLAM: bundle adjustment, multi-view reconstruction, and visual odometry
Study by time budget
Pick the path that fits the time you have before your interview.