1. Traffic Sign Recognition Using Artificial Neural Network Radi Bekker 101100

2. Motivation for ANN  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.

3. The human brain  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.

4. Artificial neural network (ANN)  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.

5. Artificial Neuron  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.

6. Training- Back Propagation-1  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.

7. Training- Back Propagation-2  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".

8. My Network  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.

9. Image Filtering  Resizing the image to size 100x100.  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.

11. Results  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.

12. Conclusions  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.