MILLIMETER WAVE IMAGING FOR VEHICLE NUMBER PLATE DETECTION

Abstract

Millimeter wave (MMW) frequency (30 GHz- 300 GHz) offers unique favourable features in contrast to other frequency bands, such as, finer resolution, biologically safe, can see through concealed object, high data rate, compact and portable systems, and can be used for short as well as long distance applications. Detection of letters/numbers from the Vehicle number plate in low light conditions is a challenging task. There is study going on to detect the numbers on the number plate using various techniques and using different electromagnetic spectrum. Therefore, this dissertation is aimed to help police and other allied departments in identifying and detecting the vehicles in low light conditions, haze, fog, smoke, dust etc. The experiment was conducted using vector network analyser (Agilent N5247A (10 MHz – 67 GHz) PNA-X), pyramidal horn antenna. (MESA. MW-HF-907A) and VNA cable (MMW-N4697F (DC to 67 GHz) – 1.85 mm), this combination is worked as SFCW radar in mono-static mode. Since letter identification and recognition are such that the capability of the system to operate in real time is of prime importance, PC is interfaced with the Vector Network Analyzer (VNA) and data acquisition and processing is carried out in real time. Firstly Delay and Sum Algorithm was implemented on the data so obtained from our experimental setup. In the conventional approach for the delay sum algorithm the setup is used in which the antenna is moved for scanning and target is fixed, but, in our experimental process we have kept the antenna location fixed and target is moved for scanning, as a result of which we use inverse phase cancellation. Finally, an Artificial Neural Network (ANN) is proposed by which we can characterize the letters. ANN is much like human brain which learns by example. It can be configured for specific problems such as prediction, pattern recognition or data classification. Backpropagation algorithm is used and studied for learning neural network. Feed forward backpropogation network type is used with Levenberg-Marquardt training function. Performance function chosen is Mean squared error. The ANN so designed is trained to six different target/letters namely: H, E, J, I, R and U. The results obtained from the neural network once trained are analogous to the desired output.