Hamza Basharat
All work

Edge inference on constrained hardware

Industrial Detection & Pose-Estimation Suite

Status
In production
Role
AI & Computer Vision Engineer
Organisation
Robionix Technologies
Year
2024-2026

The problem

Industrial vision has to run on whatever hardware is already bolted to the line - a GPU server is rarely on offer.

Outcome

mAP
92%+mAP
Throughput
15+ FPSThroughputOn constrained edge hardware.

Full-body keypoint tracking holding through movement — the pose branch of the suite.

Full-body pose keypoint tracking — the estimation technique behind the industrial detection and pose suite.

Approach

  • 01

    YOLOv8 detection plus keypoint pose estimation, tuned to hold accuracy inside an embedded compute budget.

  • 02

    Optimised the inference path for constrained edge hardware rather than assuming a datacentre GPU.

Stack

  • YOLOv8
  • ONNX
  • PyTorch
  • Embedded inference

Restricted. Client system. Metrics are shareable; the deployed footage may not be — the demo shown here is Hamza’s own.