Computer Vision Engineer

We're Hiring: Computer Vision Engineer

Location: Hubli, Karnataka

Experience: 2+ years

Domain: Computer Vision | EO & Thermal Imaging | Object Detection & Tracking | Edge AI

Freshers are not eligible. Candidates must be willing to relocate.

About this role

You will own the computer vision pipeline for real-time electro-optical systems used in defence applications.

You will work directly with live EO and thermal camera feeds and develop algorithms for detection, classification, tracking, image enhancement and camera alignment. The objective is not to demonstrate models on recorded datasets, but to make them work reliably on real hardware under changing range, motion, lighting, weather, background and thermal conditions.

Your algorithms will run on edge compute hardware and interface directly with embedded software and motion-control systems.

This is a hands-on computer vision role, not a data-science, web AI or research-only position.

What you'll be doing

Detection & Tracking

- Develop and optimise real-time object detection models for EO and thermal imagery
- Develop robust single-object and multi-object tracking pipelines
- Maintain reliable tracks through motion, partial occlusion, scale changes, clutter and temporary loss of detection
- Implement target acquisition, track initiation, track confirmation, loss detection and reacquisition logic
- Combine detector outputs with motion models and filtering techniques for smoother and more reliable tracking
- Minimise false detections and track switching under difficult backgrounds and field conditions

EO & Thermal Vision

- Work directly with live visible-spectrum and thermal camera feeds
- Develop preprocessing and image-enhancement algorithms for low-contrast, low-light and thermally challenging scenes
- Optimise detection and tracking independently for EO and thermal imagery
- Develop image registration/alignment between multiple camera channels where required
- Handle differences in resolution, field of view, frame rate and imaging characteristics between sensors
- Characterise vision performance across different ranges, target sizes and environmental conditions

Camera Geometry & Calibration

- Implement intrinsic and extrinsic camera calibration
- Develop camera-to-system alignment and boresight calibration routines
- Correct lens distortion and other geometric errors
- Implement coordinate transformations between image-space measurements and system coordinates
- Support calibration and alignment of multiple optical sensors

Model Development & Optimisation





- Train, evaluate and improve deep-learning models using field-collected datasets
- Build data pipelines for dataset preparation, augmentation and validation
- Analyse failure cases and perform hard-negative mining to systematically improve model performance
- Optimise neural networks for real-time inference on edge compute platforms
- Convert and deploy models using TensorRT, ONNX or equivalent inference frameworks
- Profile and optimise inference latency, GPU utilisation, memory usage and throughput
- Evaluate trade-offs between model accuracy, latency, compute requirement and power consumption

System Integration & Testing

- Integrate the vision pipeline with live camera streams and embedded software
- Output tracking information with sufficiently low latency for downstream real-time control
- Work with embedded software, electronics and control engineers during complete system integration
- Develop logging and visualisation tools for debugging vision performance
- Define and measure metrics including detection probability, false-alarm rate, tracking continuity, tracking error, reacquisition performance and end-to-end latency
- Participate in outdoor and field trials, analyse recorded data and improve algorithms based on actual failures
- Take algorithms from workstation development through deployment and validation on production-intent hardware

Mandatory skills and experience

- 2+ years of hands-on computer vision or perception development, with algorithms deployed on real video rather than only academic datasets
- Strong Python and working proficiency in C++
- Strong working knowledge of OpenCV
- Experience with PyTorch or an equivalent deep-learning framework
- Practical experience with real-time object detection
- Practical experience with object tracking
- Understanding of tracking concepts such as Kalman filtering, association, motion models and track management
- Experience training and evaluating deep-learning vision models
- Experience deploying models using TensorRT, ONNX Runtime or equivalent
- Understanding of camera geometry, intrinsic/extrinsic calibration and coordinate transformations




- Experience processing live camera/video streams
- Strong understanding of inference latency, frame rate, threading and real-time performance constraints
- Ability to analyse model failures systematically rather than only retrain models with more data
- Comfortable working with Linux-based development and deployment environments

Good to have

- Experience working with thermal / LWIR cameras
- Experience with EO/IR image processing and enhancement
- Experience with NVIDIA Jetson or similar edge AI platforms
- CUDA/GPU optimisation experience
- Experience with long-range or small-object detection
- Experience with low-light, low-contrast or cluttered-scene computer vision
- Experience with optical flow, feature matching, image registration or visual motion estimation
- Experience with visible-thermal image registration or multi-sensor vision
- Familiarity with GStreamer or V4L2
- Experience collecting, cleaning, annotating and maintaining real-world computer vision datasets
- Familiarity with model quantisation, pruning or other edge-inference optimisation techniques

What we will evaluate you on

We care less about how many models you have trained and more about whether you can make computer vision work reliably outside a controlled dataset.

You should be able to explain:

- Why a detector or tracker fails in a particular scene
- How you would diagnose false positives and missed detections
- How you would improve tracking when detections intermittently disappear
- How camera motion affects tracking
- How EO and thermal imagery differ from a computer vision perspective
- How you would balance accuracy against latency on constrained edge hardware
- How you would establish whether an improvement seen on a dataset actually improves field performance

Why join us

- Work on computer vision that operates on real EO and thermal hardware, not offline demonstrations
- See your algorithms progress from development datasets to integrated hardware and field trials
- Own the complete perception pipeline from data and models through optimisation and deployment
- Work directly with embedded software, electronics, controls and mechanical engineers
- Solve computer vision problems where latency, reliability and real-world performance matter

About Astr Defence

Astr Defence is an award-winning Indian defence manufacturer developing next-generation defence systems for modern battlefield requirements.

Apply: Send your profile to [email protected] with the subject "Computer Vision Engineer".

📌 Computer Vision Engineer (CV) (Hubli)
🏢 Astr Defence
📍 Hubli