Performance Comparassion SSD MobileNet and YOLOv5 on Intelligent Speed Assistance

The development of technology forces object detection algorithms to be a vital part in many aspects of technology, especially in artificial intelligence, autonomous vehicles, and computer vision. Intelligent Speed Assistance is one form of implementation of artificial intelligence in autonomous electric vehicles by using a camera as input from the sense of sight to determine the speed of the vehicle. In previous research, an ISA system was designed with the YOLOv5 backbone model and network with an average accuracy of 0.91 and mAP of 0.96 and an average computational performance of frames per second of 3 FPS. This research shows significant improvements to the ISA system by optimizing the deep learning model using the TensorRT SSD model on the Jetson Nano. Tuning training parameters, datasets, and modification of model layers successfully improved system performance. In terms of computational cost, the system achieved 8 times increase in FPS, reaching 45 FPS. In the model accuracy parameter, there was an increase in accuracy of 0.92 and mAP of 0.95. The system can detect the model with a reading speed of 34 milliseconds per image frame.