Street Object Detection Based on Faster R-CNN

Abstract

With the increased number of traffic accidents, the research and development of smart cars have been promoted. The detection of street objects has become one of the important research topics. Generic Model detection algorithm based on Convolution Neural Network(CNN) need to design the training model, while the training and testing of the model will take a lot of time. Transfer Learning is used to fine-tune the pre-trained models, using the Image task datasets of COCO, transferring a generic deep learning model to specific one with different weights and outputs. Furthermore, the CNN structure is adjusted to improve overall performance, and the street environment is trained to the special scene. We compare the results of experiments, and the results showed that the network which is fine-tuned is effective.