Computer Vision Engineer | Multi-Camera Systems

Role Summary

We seek a Computer Vision Engineer to design and implement the mirror-aware, multi-camera perception stack powering Vizhi's real-time form tracking. You will focus on:

• Mirror Mode: Combining direct view and mirror reflection to improve depth perception and handle occlusions

• Sticky Camera Mode: Detachable auxiliary camera extending coverage while preserving privacy

• Real-Time Pose Pipelines: Robust 2D/3D skeleton reconstruction under <100ms latency constraints

• Edge Optimization: Deploying CV models on smart glasses and mobile devices

You'll collaborate with VLM, Mobile ML, and Engineering teams to power intelligent coaching systems.

Key Responsibilities

1) Multi-Camera & Mirror-Aware Architecture

• Design perception systems combining wearable cameras, mirror reflections, and sticky cameras

• Evaluate and integrate 2D pose estimation models adapted for mirror environments

• Analyze model size vs. accuracy vs. latency trade-offs on target devices

2) Calibration & 3D Reconstruction

• Implement camera calibration workflows (intrinsic/extrinsic parameters, mirror plane estimation)

• Develop algorithms to detect mirrors and distinguish real body vs. reflection using geometric consistency

• Triangulate 3D joints from multi-view 2D keypoints with reprojection error validation

• Build guided calibration flows using wearable HUD alignment markers

3) Real-Time Pose & Form Analysis

• Integrate and optimize 2D pose detectors for multi-view setups under occlusions and lighting variation

• Compute biomechanical features (joint angles, ROM, timing, symmetry) for coaching models

• Implement temporal filtering to reduce jitter while maintaining responsiveness

4) Privacy-Preserving Sticky Camera & Mirror Modes

• Support transient streams with RAM-only buffering (no persistent storage)

• Incorporate surface detection checks for safe mounting on mirrors/walls

• Maintain synchronized timestamps between wearable and auxiliary cameras

• Combine direct and mirrored views to reduce self-occlusion

5) Performance Optimization & Deployment

• Profile and optimize for <100ms end-to-end latency on target devices

• Support ONNX/TFLite conversion and collaborate on hardware benchmarking

• Reduce memory/compute overhead via pruning, quantization, and efficient pipelines

6) Evaluation & Testing

• Track metrics: 2D keypoint accuracy, 3D reconstruction error, reflection misclassification rate, latency

• Build evaluation datasets for mirror-rich environments, occlusions, body types, lighting conditions

• Implement automated regression tests for mirror-aware performance

7) Cross-Functional Collaboration

• Provide APIs exposing pose keypoints, 3D skeletons, form features, and mirror diagnostics

• Work with VLM/Agent teams on multimodal data workflows

• Collaborate with Mobile ML on resource-aware deployment

• Align perception with gym use cases through Product/Trainer feedback

Required Skills & Experience

Educational Background

• Bachelor's degree in Computer Science, Engineering, Mathematics, or related field

• Master's in Computer Vision/Robotics is a plus

• Strong foundation in linear algebra, geometry, numerical methods

Core Computer Vision Knowledge

Image Processing & Analysis:

• Image operations (reading, writing, manipulation, format conversion)

• Enhancement techniques (histogram equalization, contrast adjustment, noise reduction)

• Filtering and smoothing (Gaussian, bilateral, median filters)

• Edge detection and contour analysis

Camera Geometry & Calibration:

• Camera models (pinhole geometry, intrinsic/extrinsic parameters)

• Lens distortion correction (radial and tangential)

• Camera calibration using checkerboard patterns

• Distortion correction and undistortion workflows

Multi-View Geometry & 3D Reconstruction:

• Stereo vision and depth computation

• Epipolar geometry and fundamental matrix

• Triangulation from multiple views (Direct Linear Transform)

• Stereo rectification and correspondence

• Homography estimation and image warping

• Handling degenerate cases in 3D reconstruction

Feature Detection & Matching:

• Feature descriptors (SIFT, ORB, AKAZE)

• Feature matching and outlier rejection

• Image registration and alignment

• Transformation estimation (affine, perspective)

Object & Pose Detection:

• Human pose estimation (keypoint detection, skeleton tracking)

• Object detection integration (YOLO, SSD, or similar)

• Person segmentation and tracking

• Face detection and landmark extraction

Video Processing:

• Frame capture and video I/O

• Temporal filtering and motion analysis

• Frame synchronization in multi-camera systems

• Real-time streaming and buffering

Mirror-Processing:

• Mirror optics and reflection properties

• Reflection detection and segmentation

• Mirror-corrected triangulation (3D reconstruction from direct + mirrored views)

• Handling reflection artifacts (specular highlights, partial reflections, edge cases)

Programming & Tools

Expert-level Python (3+ years):

• NumPy, SciPy for numerical computing

• OpenCV for computer vision operations

• Matplotlib/Seaborn for visualization

• Scikit-image for image processing

Proficient OpenCV (3+ years):

• Camera calibration (calibrateCamera, undistort)

• Stereo operations (stereoRectify, triangulatePoints)

• Feature detection (SIFT, ORB, BFMatcher)

• Geometric transformations and warping

• Video capture and frame handling

• DNN module for model integration

C++ Experience (1+ year):

• Reading and modifying performance-critical CV code

• Basic GPU/CUDA understanding

Version Control:

• Git workflows, code review, collaborative development

Real-Time Systems & Performance

• Experience optimizing for <100ms latency targets

• Profiling Python/C++ code (timing, memory, CPU/GPU utilization)

• Understanding GPU vs. CPU trade-offs

• Frame synchronization and temporal alignment

• Video pipeline and streaming data handling

Mathematical Foundation

• Linear Algebra: Matrix operations, SVD, QR decomposition, homogeneous coordinates

• Geometry: Projective geometry, 3D transformations, cross/dot products

• Numerical Methods: Least-squares fitting, handling ill-conditioned matrices

• Calculus: Optimization basics, gradients in geometric problems

Preferred Qualifications

OpenCV Certifications (Strongly Preferred)

We highly value candidates with OpenCV certifications demonstrating structured CV knowledge and hands-on implementation experience. Certifications validate proficiency in camera calibration, stereo vision, feature matching, pose estimation, and real-time optimization.

If you don't have certifications yet: Completing relevant OpenCV courses before applying will strengthen your candidacy.

Additional Preferred Skills

• Production CV system deployment experience

• Multi-camera systems (stereo rigs, synchronized capture)

• Real-time CV applications (robotics, AR/VR, sports analytics)

• Experience with reflective surfaces or challenging lighting

• Deep learning model integration (HRNet, YOLOv8, MediaPipe)

• GPU optimization (CUDA) and parallel processing

• Computer vision research familiarity (CVPR, ICCV, ECCV papers)

• Open-source CV contributions on GitHub

What You'll Gain

• Technical ownership of mirror-aware perception stack for patented workout guidance system

• Experience building real-time, on-device CV for smart glasses in real-world environments

• Collaboration with VLM, agent, and mobile specialists on frontier multimodal product

• Contribution to Nutpaa's patent portfolio

• Deep expertise in multi-camera systems, mirror processing, 3D reconstruction, real-time optimization

• Path to senior CV roles

MVP Success Criteria (8 Months)

1. Mirror Mode and Sticky Camera Mode validated in real gyms/homes

2. 2D/3D pose pipeline at <100ms latency on target devices

3. Reflection-handling with low misclassification and stable tracking

4. Clean APIs for VLM/agent systems

5. Evaluation suite and regression tests in CI

Work Arrangement

• Duration: 8 months (March–October 2026)

• Commitment: Full-time, 45–50 hours/week

• Location: Hybrid – Chennai or Madurai

• Post-MVP: Flexible hybrid working from Month 6+

Application Process

Email careers@nutpaa.ai with:

1. Resume highlighting:

• CV and pose estimation experience (3–4+ years)

• OpenCV certifications (include certificate numbers/links if available)

• Multi-camera, mirror/reflective environment, or AR/VR projects

• Real-time/edge deployment work

2. Portfolio:

• GitHub with CV projects (calibration, triangulation, pose estimation)

• Technical writing (blog posts, documentation)

• OpenCV-based implementations

3. Statement (~200 words):

• Interest in mirror-aware, multi-camera perception for fitness

• One complex CV project (occlusions, reflections, latency, multi-camera)

• Excitement about production CV for smart glasses

Email Subject: Computer Vision Engineer – Mirror-Aware Pose – [Your Name]

Equal Opportunity

Nutpaa is an equal opportunity employer. We value CV fundamentals, OpenCV proficiency, real-world experience, and shipping mindset over strict credentialism.

Non-traditional backgrounds welcome: Candidates without formal degrees but with demonstrable expertise (portfolio, certifications, projects) are encouraged to apply.

Questions? Email careers@nutpaa.ai