Detecting pedestrians, vehicles, and traffic lights with SSD
When a self-driving car is on a road, it surely needs to know where the lanes are and detect obstacles (including people!) that can be present on the road, and it also needs to detect traffic signs and traffic lights.
In this chapter, we will take a big step forward, as we will learn how to detect pedestrians, vehicles, and traffic lights, including the traffic light colors. We will use Carla to generate the images that we need.
Solving our task is a two-step process:
- Firstly, we will detect vehicles, pedestrians, and traffic lights (no color information), where we will use a pre-trained neural network called SSD.
- Then, we will detect the color of the traffic lights, where we will need to train a neural network starting from a pre-trained neural network called Inception v3, using a technique called transfer learning, and we will also need to collect a small dataset.
So, let's begin by using Carla...
