Traffic Sign Recognition Using Artificial Neural Network Radi

Traffic Sign Recognition Using Artificial Neural Network Radi Bekker 101100

Motivation for ANN o o o von Neumann machines are based on the processing – one processing unit, many operations in one second. Neural networks are based on the parallel architecture of animal brains-slow , parallel and complicated-good for pattern matching. Pattern matching can solve many problems to which algorithms are not exist or very complicated.

The human brain o o o Consists from 1011 neurons Neurons are connected by around 1015 connections. Neurons send impulses to each other through the connections and these impulses make the brain work. Dendrites- responsible for input. Axon- responsible for output.

Artificial neural network (ANN) o o o Network is constructed from artificial neuron layers. There is input and output layers and any number of hidden (internal) layers. Each neuron in one layer is connected to every neuron in the next layer.

Artificial Neuron o o o Many inputs like dendrites. One output like axon. Each neuron receives a signal from the neurons in the previous layer. The weighted inputs are summed, and passed through a limiting function which scales the output to a fixed range of values. The output of the limiter is then broadcast to all of the neurons in the next layer.

Training- Back Propagation-1 o o o The most common learning algorithm is called Back Propagation (BP). A BP network learns by example, that is, we must provide a learning set that consists of some input examples and the known-correct output for each case. This method adjusts the weights between the neurons to solve a particular problem. The BP learning process works in small iterative steps: one of the example cases is applied to the network, and the network produces some output based on the current state of it's synaptic weights. This output is compared to the known-good output, and a mean-squared error signal is calculated.

Training- Back Propagation-2 o o The error value is then propagated backwards through the network, and small changes are made to the weights in each layer. The whole process is repeated for each of the example cases, then back to the first case again, and so on. The cycle is repeated until the overall error value drops below some pre-determined threshold. At this point we say that the network has learned the problem "well enough".

My Network o o Input layer-10, 000 neurons. Hidden layers-3 hidden layers with 10 neurons each. Output layer-16 neurons for 16 traffic signs. Training- network trained for 2000 cycles.

Image Filtering o o Resizing the image to size 100 x 100. Turning the image to black and white. Rescaling the matrix image to numbers between 0 and 1. Constructing a 10, 000 sized vector from the columns of the image matrix.

Results o o Good results for trained images Bad results for real picture images. When the network was constructed to identify 5 images- better results was achieved. Contrast and brightness adjustments in some cases contributed to sign correct recognition.

Conclusions o o o ANN is good for small problems and networks. ANN is bad for big networks. Bigger network –more training time needed. Hard to find out good network configurations. ANN is a good method for solving hard computational problems. More research on human brain could be helpful in constructing better ANN.