Assignment on Car type classification

Introduction of Car type classification

In this report, there is some comprehensive information about the system that can easily identify the type of car through the help of a picture. For that case, the user will input a picture of a car and it will go towards the system and the output will evaluate the type of that car. The whole system contains a complete dataset of modern cars or fast-moving cars. This system is extremely helpful in high speed chases on the highway in speed cameras.

High-level working of the system of Car type classification

In this section, there is a complete working of the whole system. It will start from the data set, for that case, the dataset will contain all images of the modern car in the system. This shows that the output image will be compared with all data set images and find out the type of that particular car. For that case, the camera must be equipped with a digital PAL resolution for clear images. The dataset of the car will consist of more than 6000 cars. The images will be of the same dimensions (Huttunen, Yancheshmeh, & Chen, 2016).

Classification of Car type classification

Now in this step, there is complete information about the system that will classify the type of car. The system for detection is deep neural networks. For that case, there are about different factors that are contributing to its success, which include a graphing processing unit and the advancement in the neural network community. Both of these factors are involved in enabling the training network with a large number of stack layers.

Then, it can be noted that a large scale image classification record that contains the ImageNet database will play an important role during recognition. Then after this in the next step there is a need to define hyperparameters to the system. This is because it is used for searching. It can be noted that the whole system consists of 4 convolutional layers and it also contains two dense layers. Due to this the input image is then classified into a square shape with the dimension of 60-160 (Ma & Andreasson, 2007).

Then after this, for stochastic gradient backpropagation, a crucial parameter is used. This parameter contains geometric distribution between Then after this a network that contains 50 random selected hyper parameter will play a role.

There are about 5 convolutional layers in the car recognition. For that case, the first layer will map the three-channel 96*96 input of a car image into 32 feature maps. These maps are pooled up to 48*48 resolution. Now after this, the second convolutional layer will also producing 32 feature maps. Then these maps are again down sampled with 24*24 max-pooling. Moreover, after the convolutional layer there are more two layer that are connected with each other with 100 nodes. Now in the last step, the output layer will map the 100 features on the last layer with the help of softmax operator. It will show the clear image of the car and also it is classified in a proper way.  Moreover, in every layer there is complete combination of a rectified linear unit and a dropout regularizer that are helping in classification with perfection.

Figure 1: Structure of the proposed network

The above figure is the brief information about the whole system. This figure is defining the complete working of the system through deep neural networks. It can be seen from the figure that the input image is taken and the output is classifying it into a van.

The main benefit of this system is that it will enable the classifier to lever high level structures in the image. Moreover, there is also one drawback in the system and it is related to the hyper parameters of the network. This shows that there is need to define these parameters to the network (Yang, Luo, Loy, & Tang., 2015).

The vehicle classification process is a critical process. Kul et al (2017) proposed a model technique for the classification and detection of cars. The framework mainly consist of input frames, background subtraction, feature extraction, classification and other detection results. The approached worked well for the target detection and feature based object detection refer to the algorithm to detect the objects (Kul, Eken, & Sayar, 2017). The research conducted by Yousaf et al (2012) was mainly about the vehicle classification and emerged as a significant field of the study. The system provides applications like security system, accident prevention, congestion avoidance, and algorithms to implement the classification. The research proposed Hybrid Dynamic Bayesian Network (HDBN) classification algorithm for the better estimation of cars, features, classification, and selection. The research extracts the rear view of information such as distance between the wheels of cars and height of wheels (Yousaf, Iftikhar, & Javed, 2012).

Expected input and output of the system

From the above system it can be noted that the expected input of the system is the image. But it can be noted that the input image must be 96*96 dimension image and it must be taken through the digital camera. The input image also contain 3 important channels. On the other hand, the output of the system will give classified the image into the different category of the cars. This shows that the output will be defined in 4 likelihoods.

Objectives for the system of Car type classification

·         To evaluate the classification of the type of car

·         To check how the system will work according to the required input

·         To analyse brief working of the car type classification system

·         To analyse how the required output is gained through the system

·         To evaluate positive and negative points of the system.

Conclusion of Car type classification

Summing up all the discussion from above, it is concluded that car type classification system is extremely effective in high speed chasing around the world. In this report there is complete information about the whole system. This system is using deep neural network. The main benefit of this system is that it will enable the classifier to lever high level structures in the image. In this report, the working of whole system is presented in a proper way. Moreover, the main objective of the system is also evaluated. The working of the system is also explained through the help of figure.

References of Car type classification

Huttunen, H., Yancheshmeh, F. S., & Chen, K. (2016). Car type recognition with deep neural networks. In 2016 IEEE Intelligent Vehicles Symposium (IV),.

Kul, S., Eken, S., & Sayar, A. (2017). A Concise Review on Vehicle Detection and Classification. Conference on Engineering and Technology, 01(05), 01-10.

Ma, X., & Andreasson, I. (2007). Behavior measurement, analysis, and regime classification in car following. IEEE transactions on intelligent transportation systems 8, .

Yang, L., Luo, P., Loy, C. C., & Tang., X. (2015). A large-scale car dataset for fine-grained categorization and verification. In Proceedings of the IEEE conference on computer vision and pattern recognition,.

Yousaf, K., Iftikhar, A., & Javed, A. (2012). Comparative Analysis of Automatic Vehicle Classification Techniques: A Survey. Automatic Vehicle, 36(18), 01-10.