Computer Vision Interview 20 essential Q&A Updated 2026
Autonomous

Autonomous Vehicles (CV): 20 Essential Q&A

Multi-sensor perception, real-time constraints, and validation for self-driving stacks.

~12 min read 20 questions Advanced
lanesdetectionfusionLiDAR
Quick Navigation
1. Perception stack overview
2. Camera vs LiDAR vs radar
3. Sensor fusion
4. Lane detection
5. Semantic segmentation
6. Monocular depth
7. Object detection
8. Tracking & prediction
9. HD maps
10. Calibration
11. Weather / night
12. Functional safety
13. Simulation
14. Long-tail objects
15. Occlusion
16. Latency budgets
17. Bird’s-eye view
18. ODD definition
19. Annotation cost
20. End-to-end trends
1 What does perception do in AVs? ⚡ easy
Answer: Estimate drivable space, lanes, traffic actors, signs, and hazards from sensors to support planning and control.
2 Camera vs LiDAR vs radar? 📊 medium
Answer: Camera: rich semantics, cheap; LiDAR: accurate range, weather limits; radar: velocity, robust weather—stacks often fuse all three.
3 Sensor fusion levels? 🔥 hard
Answer: Early (raw/feature), object-level, late decision fusion—trade calibration complexity vs robustness to single-sensor failure.
4 Lane detection? 📊 medium
Answer: Segmentation masks, polynomial fits, or transformer lanes in BEV—must handle markings, merges, and construction zones.
5 Segmentation use? 📊 medium
Answer: Drivable area, road vs sidewalk, freespace for parking—often multi-class at high resolution with temporal smoothing.
6 Monocular depth? 📊 medium
Answer: Supplement LiDAR in camera-only tiers or dense depth for fusion—learned depth can fail on unseen textures.
7 Detection classes? ⚡ easy
Answer: Vehicles, pedestrians, cyclists, traffic lights/signs—need range, velocity hooks for tracker and planner.
8 Tracking role? 📊 medium
Answer: Maintain stable IDs, smooth boxes, predict future motion—critical for collision avoidance and behavior prediction.
9 HD maps? 🔥 hard
Answer: Centimeter lane geometry, semantics—anchor localization; mapless stacks push more burden onto online perception.
10 Calibration? 📊 medium
Answer: Extrinsics drift, vibration—online self-calibration vs factory; bad cal breaks fusion and projection.
11 Weather / night? 📊 medium
Answer: Sensor degradation, glare, spray—domain adaptation, multi-sensor redundancy, conservative ODD restrictions.
12 Functional safety (concept)? 🔥 hard
Answer: ISO 26262 mindset: fault detection, redundancy, validated perception uncertainty for ASIL-rated paths—not just model accuracy.
13 Simulation? 📊 medium
Answer: CARLA, NVIDIA DRIVE Sim—scale rare scenarios; sim-to-real gap remains a research and validation topic.
14 Long-tail objects? 📊 medium
Answer: Debris, animals, unusual vehicles—need active learning, fleet logging, and conservative planner reactions.
15 Occlusion? ⚡ easy
Answer: Pedestrians between cars—temporal reasoning, bird’s-eye fusion, and prediction to “see” briefly hidden actors.
16 Latency budgets? 📊 medium
Answer: End-to-end perception often tens of ms—tensorRT, sparse models, ROI processing; planner assumes aged observations.
17 Bird’s-eye view models? 🔥 hard
Answer: Lift image features to 3D/BEV grid (LSS, transformers) for consistent multi-camera reasoning—popular in modern detectors. 
# BEV: lift 2D features to bird's-eye grid for fusion
18 What is ODD? ⚡ easy
Answer: Operational design domain—where the system is validated to operate; leaving ODD requires disengagement or human takeover.
19 Annotation? 📊 medium
Answer: LiDAR cuboids, polyline lanes, radar association—expensive; weak labels and self-supervision reduce cost.
20 End-to-end driving? 🔥 hard
Answer: Direct sensor→control learning challenges interpretability and safety case—hybrid stacks dominate production today.

AV Perception Cheat Sheet

Sensors
  • Cam + LiDAR + radar
Tasks
  • Lanes / det / track
Ops
  • ODD + latency

💡 Pro tip: Fusion + temporal tracking; never ignore ODD and safety case.

Full tutorial track

Go deeper with the matching tutorial chapter and code examples.

Autonomous Vehicles Tutorial
Previous Next