Performance Analysis of SSD and Faster RCNN Multi-class Object Detection Model for Autonomous Driving Vehicle Research Using CARLA Simulator

Autonomous vehicle research has grown exponentially over the years with researchers working on different object detection algorithms to realize safe and competent self-driving systems while legal authorities are simultaneously looking into the ways of mitigating the risks posed by fully autonomous vehicles. These advancements can result in a much safer commuting environment, reduced accidents and also eliminate the necessity for human driving. Recent developments in the field show that object detection models combined with an on-vehicle camera module provides more robustness and accuracy than other methods such as LiDAR or RADAR. This paper proposes two object detection algorithms, SSD and Faster RCNN for autonomous driving applications through various performance parameters. CARLA Simulator was used to generate synthetic data to train and test the models. Results shows that that Faster-RCNN was found to have a mean Average Precision (mAP) value of 94.32% while SSD has a mAP of 88.998%. However, SSD had a speed of 30 ms/image while Faster-RCNN had a speed of 106 ms/image. Taking into consideration the real-time and speed constraints in autonomous driving, it was inferred that the SSD algorithm is much better suited for this problem as the difference in accuracy between the models was relatively lesser compared to the difference in computation speeds.