Recognition of vehicle-logo based on faster-RCNN

Abstract

Vehicle-logo recognition, consisting of vehicle-logo location and its classification, is an important application of object detection in intelligent transportation. In this paper, we adopt the strategy of integrating Faster-RCNN model with two different convolutional neural networks (VGG-16 and ResNet-50) respectively and prepare a vehicle-logo images dataset containing 4000 vehicle images with different angles, backgrounds and resolutions for 8 different vehicle logos. In our experiments, the better mean Average Precision result of 94.33% is achieved in spite of the small proportion, huge intra-class variability and complex external environment of vehicle logos in the images, which shows that the methods based on Faster-RCNN can be used to recognize vehicle logos of road-monitoring vehicles and have good robustness. Integrating Faster-RCNN model with VGG-16 is better than ResNet-50 in the dataset we prepare, which illustrates the deeper network may not be the better for different recognition tasks with different amount of data.