← Back to Careers
air · software

Perception / Computer Vision Engineer

$160k – $265k

Build the perception stack — detection, tracking, depth estimation — that lets autonomous systems understand what they see.

Why it matters

Perception is the #1 bottleneck for autonomy. Modern CV/ML expertise applied to noisy, real-world drone imagery pays top-of-market at every prime and startup.

Skills at a glance
1

CV & deep learning foundations

8 months

Classical CV + modern deep learning. You need both — neural networks don’t free you from knowing what a Jacobian is.

Course
🎓 CourseDeep Learning Specialization· Coursera
essential60hintermediate$49

Andrew Ng’s cornerstone course — still the fastest DL ramp-up.

Learn →
🎓 CourseCS231n — Deep Learning for CV· Stanford
importantFree

Stanford’s open CV course — assignments are the real ramp.

Learn →
Skill
🛠 SkillPyTorch for vision· PyTorch docs
essentialFree

Nearly all research + production CV runs on PyTorch.

Learn →
Knowledge
📖 KnowledgeMultiple View Geometry (Hartley & Zisserman)· Amazon
important

Geometry still matters — SLAM, VIO, SfM all come from this book.

Learn →
2

Detection, tracking, 3D perception

6 months

Moving from image-level ML to full 3D world understanding.

Skill
🛠 SkillObject detection (YOLO / DETR)· Ultralytics
essentialFree

The workhorse perception task on every UAV.

Learn →
🛠 SkillVisual-Inertial Odometry (VIO)· OpenVINS
essentialFree

GPS-denied navigation starts here.

Learn →
🛠 SkillNeRF / 3D Gaussian Splatting
recommendedFree

Neural scene reconstruction is starting to replace classical SfM/MVS for offline mapping, synthetic-data generation, and post-mission analysis.

🛠 SkillFoundation models for perception (SAM, GroundingDINO, CLIP)
importantFree

Defense ISR is moving from task-specific CNNs to fine-tuned foundation models. Few-shot adaptation on small operator-labeled datasets is the new perception workflow.

Course
🎓 CourseSelf-Driving Cars Specialization· Coursera
important80hintermediate$49

Perception + sensor fusion patterns that port straight to aerial.

Learn →
Tool
🔧 ToolOpenCV + ROS 2 integration· ROS 2 docs
importantFree

Every perception node in a flight stack speaks ROS image msgs.

Learn →
3

Deployment on embedded

6 months

A model that only runs on a DGX is useless on a drone. This stage is about making it fit + fly.

Skill
🛠 SkillTensorRT / ONNX Runtime on Jetson· NVIDIA
essentialFree

Most airborne inference runs on Jetson via TensorRT.

Learn →
🛠 SkillModel quantization + pruning· PyTorch
importantFree

INT8 + pruning is how a ResNet-50 fits into a 15W payload budget.

Learn →
Certification
📜 CertificationNVIDIA Jetson Professional· NVIDIA
recommendedFree

Vendor cert that hiring managers recognize.

Learn →
Course
🎓 CourseEdge AI with TensorRT· NVIDIA DLI
importantFree

NVIDIA DLI course — hands-on Jetson deployment.

Learn →