Edge Device Aerial Detection - 03/02/2026 12:21 EST
Budget: ₹12,500 – ₹37,500 INR
I need a compact, fast object-detection model that runs directly on an edge board (Jetson Nano, Raspberry Pi, Coral or similar) and processes aerial images from drones. The immediate application is surveillance, yet the solution should stay flexible enough to be reused later in agriculture or disaster-response scenarios.
Source imagery will contain a mix of people, cars, buses, bicycles and assorted infrastructure. The model must be especially reliable at spotting people and critical infrastructure elements, while still recognising the wider vehicle classes.I am open to any justified architecture (YOLOv8, MobileNet-SSD, EfficientDet-Lite, or superior alternatives)—provided it outperforms YOLOv9c and similar models while delivering real-time inference on edge devices once quantized or otherwise optimized.
Deliverables
• Edge-ready model file (TensorRT, TFLite or ONNX)
• Python inference script with a one-command launch and clear README
• Evaluation report showing mAP, FPS and a short demo clip on a held-out aerial set
We can stage the work through prototype, optimisation and final hand-off milestones. If any dataset you plan to use carries licence fees, flag that up front.
Source imagery will contain a mix of people, cars, buses, bicycles and assorted infrastructure. The model must be especially reliable at spotting people and critical infrastructure elements, while still recognising the wider vehicle classes.I am open to any justified architecture (YOLOv8, MobileNet-SSD, EfficientDet-Lite, or superior alternatives)—provided it outperforms YOLOv9c and similar models while delivering real-time inference on edge devices once quantized or otherwise optimized.
Deliverables
• Edge-ready model file (TensorRT, TFLite or ONNX)
• Python inference script with a one-command launch and clear README
• Evaluation report showing mAP, FPS and a short demo clip on a held-out aerial set
We can stage the work through prototype, optimisation and final hand-off milestones. If any dataset you plan to use carries licence fees, flag that up front.