Hamed Habibi Aghdam

A practical approach for detection and classification of traffic signs using Convolutional Neural Networks

Automatic detection and classification of traffic signs is an important task in smart and autonomous cars. Convolutional Neural Networks has shown a great success in classification of traffic signs and they have surpassed human performance on a challenging dataset called the German Traffic Sign Benchmark. However, these ConvNets suffer from two important issues. They are not computationally suitable for real-time applications in practice. Moreover, they cannot be used for detecting traffic signs for the same reason. In this paper, we propose a lightweight and accurate ConvNet for detecting traffic signs and explain how to implement the sliding window technique within the ConvNet using dilated convolutions. Then, we further optimize our previously proposed real-time ConvNet for the task of traffic sign classification and make it faster and more accurate. Our experiments on the German Traffic Sign Benchmark datasets show that the detection ConvNet locates the traffic signs with average precision equal to 99.89 % . Using our sliding window implementation, it is possible to process 37.72 high-resolution images per second in a multi-scale fashion and locate traffic signs. Moreover, single ConvNet proposed for the task of classification is able to classify 99.55 % of the test samples, correctly. Finally, our stability analysis reveals that the ConvNet is tolerant against Gaussian noise when ź < 10 . Implementing the scanning window technique on ConvNets using dilated convolutions.Proposing a highly optimized and accurate ConvNet for classification of traffic signs.Analyzing stability of the proposed ConvNet on noisy images.

Recognizing Traffic Signs Using a Practical Deep Neural Network

Convolutional Neural Networks (CNNs) surpassed the human performance on the German Traffic Sign Benchmark competition. Both the winner and the runner-up teams trained CNNs to recognize 43 traffic signs. However, both networks are not computationally efficient since they have many free parameters and they use highly computational activation functions. In this paper, we propose a new architecture that reduces the number of the parameters \(27\%\) and \(22\%\) compared with the two networks. Furthermore, our network uses Leaky Rectified Linear Units (Leaky ReLU) activation function. Compared with 10 multiplications in the hyperbolic tangent and rectified sigmoid activation functions utilized in the two networks, Leaky ReLU needs only one multiplication which makes it computationally much more efficient than the two other functions. Our experiments on the German Traffic Sign Benchmark dataset shows \(0.6\%\) improvement on the best reported classification accuracy while it reduces the overall number of parameters and the number of multiplications \(85\%\) and \(88\%\), respectively, compared with the winner network in the competition. Finally, we inspect the behaviour of the network by visualizing the classification score as a function of partial occlusion. The visualization shows that our CNN learns the pictograph of the signs and it ignores the shape and color information.

A Unified Framework for Coarse-to-Fine Recognition of Traffic Signs using Bayesian Network and Visual Attributes

Recently, impressive results have been reported for recognizing the traffic signs. Yet, they are still far from the real-world applications. To the best of our knowledge, all methods in the literature have focused on numerical results rather than applicability. First, they are not able to deal with novel inputs such as the false-positive results of the detection module. In other words, if the input of these methods is a non-traffic sign image, they will classify it into one of the traffic sign classes. Second, adding a new sign to the system requires retraining the whole system. In this paper, we propose a coarse-to-fine method using visual attributes that is easily scalable and, importantly, it is able to detect the novel inputs and transfer its knowledge to the newly observed sample. To correct the misclassified attributes, we build a Bayesian network considering the dependency between the attributes and find their most probable explanation using the observations. Experimental results on the benchmark dataset indicates that our method is able to outperform the state-of-art methods and it also possesses three important properties of novelty detection, scalability and providing semantic information.

Analyzing the Stability of Convolutional Neural Networks against Image Degradation

Understanding the underlying process of Convolutional Neural Networks (ConvNets) is usually done through visualization techniques. However, these techniques do not provide accurate information about the stability of ConvNets. In this paper, our aim is to analyze the stability of ConvNets through different techniques. First, we propose a new method for finding the minimum noisy image which is located in the minimum distance from the decision boundary but it is misclassified by its ConvNet. Second, we exploratorly and quanitatively analyze the stability of the ConvNets trained on the CIFAR10, the MNIST and the GTSRB datasets. We observe that the ConvNets might make mistakes by adding a Gaussian noise with σ = 1 (barely perceivable by human eyes) to the clean image. This suggests that the inter-class margin of the feature space obtained from a ConvNet is slim. Our second founding is that augmenting the clean dataset with many noisy images does not increase the inter-class margin. Consequently, a ConvNet trained on a dataset augmented with noisy images might incorrectly classify the images degraded with a low magnitude noise. The third founding reveals that even though an ensemble improves the stability, its performance is considerably reduced by a noisy dataset.

A Practical and Highly Optimized Convolutional Neural Network for Classifying Traffic Signs in Real-Time

Classifying traffic signs is an indispensable part of Advanced Driver Assistant Systems. This 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 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

Toward an optimal convolutional neural network for traffic sign recognition

73\,\%

Training a Mentee Network by Transferring Knowledge from a Mentor Network

73% compared with two state-of-art ensemble of ConvNets. In addition, our ensemble is