Object Detection for Self-Driving Cars using Computer Vision and Deep Learning

Objective:
The objective of this project was to detect various objects in the scene from an FPV video stream of a vehicle. The ultimate goal was to accurately annotate most of the objects within a frame using a pretrained object detection model.

Implementation:
The pretrained neural network model chosen for this project was the Single-Shot Multi-Box Detector (SSD) Mobilenet trained on the COCO dataset. The implemented pipeline passes each frame through the model, which then returns a bounding box along with the class probabilities for every detected object in the scene. The bounding boxes are filtered by thresholding their respective confidence scores, and resulting classifications are reported accordingly. Finally, each frame is annotated with a colour-coded box bounding every detected object, along with the respective class labels.