Vehicle Violation Detection Using Faster R-CNN and SPP-Net in Bike Lanes

Clients

Techstack

Purpose

Motorists routinely stray into bike lanes, putting cyclists at real risk, and enforcement cannot scale on human eyes alone. This study set out to find which computer-vision approach could spot those intrusions accurately and reliably enough to underpin automated enforcement.

Description

We implemented two competing deep-learning detection models and ran them head-to-head on real bike-lane footage, measuring not just accuracy but how each held up as the data grew and how fast and stably it ran. The comparison produced a grounded recommendation for which approach an automated monitoring system should be built on. Rigorously benchmarking vision models against a real safety problem is the kind of applied AI research we take on.