One of the biggest misunderstanding among students and outsider robotic learners is that Object Detection = Self-Driving Cars.
​
​Somehow, this misconception keeps surviving, and many spend weeks learning all the details about the latest version of YOLO. In reality, self-driving cars are NOT using object detectors like YOLO as their main perception pipeline.

It can be a tool, that is later fused with point clouds, or used for region filtering, or other... but rarely the end algorithm.
​
​Instead, companies focus on Object Tracking. Knowing that an object is at pixel (300,300) is USELESS to a self-driving car. But knowing that the car is crossing our lane is useful.
​
​Object Tracking turns detection into a useful feature.