2D Object Detection
The 2D object detection algorithms can be broadly grouped into the following two types:
- Single stage detector — Yolo and SSD.
- Two stage detector — RCNN, Fast RCNN and Faster RCNN.
The difference between the two is that in the two stage detectors, the first stage uses region proposal networks to generate regions of interest and the second stage uses these regions of interest for object classification and bounding box regression. On the other hand, a single stage detector uses the input image to directly learn the class wise probability and bounding box coordinates. Thus these architectures treat the object detection as a simple regression problem and thus are faster but less accurate.
3D Object Detection
Both camera based and lidar based approaches have been used in literature. Lidar data has been proven to be a better alternative achieving higher accuracy than camera based approaches. The challenge with using lidar data is that it produces data in the form of point clouds which have millions of points thus increasing the computational cost and processing time. For autonomous vehicles to work, it is very important for the perception component to detect the real world objects with both high accuracy and fast inference.