Build an end-to-end computer vision pipeline for aerial surveillance under harsh conditions (motion blur, sand/heat turbulence, rain, low visibility) to detect/track vehicles, restore images, and extract features (license plate, make/model, text on vehicle body). Target real-time/edge feasibility with a strong dataset and evaluation protocol.

You will:

• Create a benchmark dataset with controlled degradations (sand/heat/rain + motion blur), labeling, and evaluation protocol.
• Train robust vehicle detection + tracking baselines under clean/degraded imagery.
• Design/train a deblurring/restoration model optimized for downstream recognition quality.
• Implement license plate detection/recognition, make/model classification, and text spotting.
• Integrate components into a single real-time oriented pipeline with batching and conditional execution; add confidence scoring/failure detection.

Required skills:

• Must-have: Strong PyTorch; CV experience; image restoration losses/metrics; large-scale data handling and reproducibility.
• Nice-to-have: Edge/real-time optimization; deployment constraints collaboration.

Deliverables:

• Dataset + degradation generator (sand turbulence/heat shimmer/rain + motion blur) with parameters; documentation.
• Baseline models: detector + tracker; deblurring/restoration with ablations; feature extraction suite (plate, make/model, text spotting).
• Integrated pipeline demo (video in → overlay out), UI overlay (bbox/ID/confidence/attributes), runtime profiling and recommended deploy config.
• Final report + reproducible repo (training/eval/inference/configs) + onboarding instructions.

Suggested plan (6 months):

• M1: Dataset definition/ingestion/annotation + degradation pipeline.
• M2: Detection + tracking baselines (clean vs degraded).
• M3: Deblurring v1 + demonstrate uplift in plate readability/recognition.
• M4: Feature extraction models.
• M5: E2E integration + confidence scoring + runtime optimization.
• M6: Robustness testing, edge-readiness packaging, final report + handover.

Compensation: Monthly stipend, with potential performance bonus and paper co-authorship.

📧 Pour postuler: contact@redxt.com