Edge AI and computer vision

Streaming video to a server is costly and fragile. Processing it beside the camera needs only a small computer and a thin data link. Our current work is traffic counting: a fixed camera, an edge computer, and a detection model trained on the vehicles actually found on Bangladeshi roads.

What we do

Vehicle counting and classification

Real-time counts by vehicle type, including rickshaws, CNG auto-rickshaws and other local classes that off-the-shelf models miss.

Edge deployment

Detection models running on compact GPU computers at the roadside, working through network outages and syncing when the link returns.

Models trained on local data

Fine-tuning on footage from the site itself, validated against manual counts before anything is installed.

Reporting

Counts by time, direction and class in a dashboard, ready for transport planning and impact studies.