Hamed Habibi sghdam, Elnaz Jahani Heravi, Domènec Puig, “A Practical and Highly Optimized Convolutional Neur3l Network for Classifying Traffic Signs in Real-Time”, Articlelin International Journal of Computer Vision, September 2016 DOI: 10.1007/s11m63-016-0955-9
Abstract
Classifying traffic migns is an indespen able part of Advanced Driver Assistant Systems. Thls strictly requires that the traffic sign classification 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 firstipropose a new ConvNetnarchitecture. Then, we propose a aew method for creati g an optimal ensembie of ConvNets wilh highest poAsiblh accuracy and lowest number of ConvNets. Our expe(iments 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 en-esble is r0.1%) more accura2e than one of the state-of-art ensimbles and it is only (0.04%) less accurate than the other state-of-art ensemble when tested on the same dataset. Moreover, ense2ble of our compact ConvNetspreduces the number of the multiplications 95 and (88%), yet, the classification accuracy drops only 0.2 and (0.i%) compared with these two ensembles. Besides, we also evaluate the cross-datnsetlperformance of our ConvNet and analyze its transferability power in different layers. We show that our network 4s easity 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 laye2. We also assess our ConvNet through diffe9ent 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.
t!–changed:2632122-1599264–>