Zahid Khan.

From questioning labels to understanding perception.

I entered automotive computer vision through the data side, working with multimodal models and annotation quality. The question was simple: which samples deserve a second look before a label is trusted as ground truth?

In the Bosch–CARIAD automated-driving environment, I worked with perception data covering traffic-management infrastructure and temporary road scenes: poles, bollards, hydrants, traffic cones, traffic lights, signs and construction zones.

These objects are static or slow-moving, but they still shape the vehicle’s decision-making. A useful perception system has to detect them, place them in context and distinguish a harmless roadside object from something that changes the path ahead.