Recognizing traffic signs is an indispensable part of Advanced Driver Assistant Systems. This strictly requires that the traffic sign recognition model accurately classifies the images and consumes as few CPU cycles as possible to immediately release the CPU for other tasks. In this paper, we first propose a new ConvNet architecture. Then, we propose a new method for creating an optimal ensemble of ConvNets with highest possible accuracy and lowest number of ConvNets. Our experiments show that the ensemble of our proposed ConvNets (the ensemble is also constructed using our method) reduces the number of arithmetic operations 88% and 73% compared with two state-of-art ensemble of ConvNets. In addition, our ensemble is 0.1% more accurate than one of the state-of-art ensembles and it is only 0.04% less accurate than the other state-of-art ensemble when tested on the same dataset. Moreover, ensemble of our compact ConvNet reduces the number of the multiplications 95% and 88%, yet, the classification accuracy drops only 0.2% and 0.4% compared with these two ensembles. Besides, we also evaluate the cross-dataset performance of our ConvNet and analyze its transferability power in different layers. We show that our network is easily scalable to new datasets with much more number of traffic sign classes and it only needs to fine-tune the weights starting from the last convolution layer. We also assess our ConvNet through different visualization techniques. Besides, we propose a new method for finding the minimum additive noise which causes the network to incorrectly classify the image by minimum difference compared with the highest score in the loss vector.
#Requirements
- You must have caffe [https://github.com/BVLC/caffe] installed on your system.
- If you want to train the ConvNet from sratch you may need to download the original dataset from [http://benchmark.ini.rub.de/].
#How to use it The ConvNet architecture as well as its solver are available in the model folder. Also, you can find the pretrained models in the caffemodel folder.