Research On Vehicle Image Segmentation And Recognition In Complex Environment

The world development and the increasing quality of life have led to an exponential increase of the number of cars used in every city.To guarantee the security of the cars in the parking and on the roads,the security agents are limited on their own.To help in the security and assistance several object detection and recognition systems are being used.These systems can be adapted to different situations but all have some limits.These limits are mainly due to the environment and the difficulty for the systems to detect distinctly different objects.Convolutional Neural Network(CNN)is one of those different method uses.Based on human and animal's brain functioning,the CNN through a training process learns different features and updates its parameters.A CNN can make a classification independently of pose,the scale,the illumination,the conformation and clutter.These abilities enable the CNN to classify complex images with a high accuracy.The Fuzzy C-means fuzzy basically used for clustering,has an accurate result while applied to image segmentation..In this thesis the idea is to be able to specify the detection to cars only.This will be done through the Convolutional Neural Network CNN's training with regard to a suitable choice of the architecture of the network and the batch-size.The different images are then classified through the network.The training and the testing of the network were made by a bank of images preprocessed and adapted to the CNN.The fuzzy c-means clustering method with is abilities to put the pixels into clusters,is used to do the segmentation of the classified image and point out the target.