Analysis Of Various Object Detection Techniques for Self-Driving Cars

Image Processing techniques like object detection, recognition and segmentation are being integrated into a wide range of applications. With the rise in integration, there is a rise in the number of different approaches to these techniques. Some techniques provide high accuracy by requiring more computation time (train or test), while some are very efficient for low end computational devices with a compromise on accuracy. There are pre-trained models that can be further trained for a specific domain using transfer learning. Other approaches include using trained models from scratch or a combination of pre-trained models along with custom models on top of it. The purpose of this paper is to test various approaches (CenterNet, EfficientNet, custom Deep-Learning models - FRCNN, SSD) in order to find a model with the accuracy-computation trade-off. The models will be trained on a customized dataset for the purpose of self-driving cars.