Surveillance cameras along urban roads capture vast amounts of unlabeled video that could be used to monitor public transport, but sorting the vehicles into meaningful categories typically requires manual annotation. A new paper on arXiv proposes an unsupervised clustering approach to address this bottleneck.
The method represents each vehicle using deep morphological features and color histograms, then groups them into clusters without relying on labeled examples. This could reduce the need for human effort in building vehicle classification systems for transit monitoring.
The abstract does not include experimental details or results, so the effectiveness of the approach is not yet reported. Still, the work points toward a practical way to leverage existing surveillance infrastructure for automated public transport oversight.