Vijendran G. Venkoparao

UWB based oil quality detection

In this paper we discuss an approach for fuel quality detection using Ultra-wideband (UWB) radar. The experimental setup consists of an ultrashort monocycle pulse transmitter (30 ps), an UWB sampling receiver and UWB antennae (3.1–10.6 GHz). One dimensional Finite Difference Time Domain (FDTD) model was used to model the electromagnetic wave propagation and assess the feasibility of oil quality detection using UWB RADAR. Performance of the UWB RADAR was analyzed through detailed experiments. The adulteration of gasoline and high speed diesel with kerosene was increased from 1% to 12% in steps. Using time of flight analysis the variation in adulteration of gasoline was detected from 1%. Adulteration in high speed diesel was detected using frequency domain analysis. Further, the level of water in the oil storage tank below the diesel or gasoline level can be accurately estimated. Detailed experiments with multiple batches of gasoline and high speed diesel were conducted. We also demonstrate that the adulteration can be detected across different batches of fuel.

Graphic Symbol Recognition Using Auto Associative Neural Network Model

Symbol recognition is a well-known problem in the field of graphics. A symbol can be defined as a structure within document that has a particular meaning in the context of the application. Due to their representational power, graph structures are usually used to represent line drawings images.An accurate vectorization constitutes a first approach to solve this goal. But vectorization only gives the segments constituting the document and their geometrical attributes.Interpreting a document such as P&ID (Process & Instrumentation)diagram requires an additional stage viz. recognition of symbols in terms of its shape. Usually a P&ID diagram contain several types of elements, symbols and structural connectivity. For those symbols that can be defined by a prototype pattern, we propose an iterative learning strategy based on Hopfield model to learn the symbols, for subsequent recognition in the P&ID diagram. In a typical shape recognition problem one has to account for transformation invariance. Here the transformation invariance is circumvented by using an iterative learning approach which can learn symbols with high degree of correlation.

Multiple appearance models for face tracking in surveillance videos

Face tracking is a key component for automated video surveillance systems. It supports and enhances tasks such as face recognition and video indexing. Face tracking in surveillance scenarios is a challenging problem due to ambient illumination variations, face pose changes, occlusions, and background clutter. We present an algorithm for tracking faces in surveillance video based on a particle filter mechanism using multiple appearance models for robust representation of the face. We propose color based appearance model complemented by an edge based appearance model using the Difference of Gaussian (DOG) filters. We demonstrate that combined appearance models are more robust in handling the face and scene variations than a single appearance model. For example, color template appearance model is better in handling pose variations but they deteriorate against illumination variations. Similarly, an edge based model is robust in handling illumination variations but they fail in handling substantial pose changes. Hence, a combined model is more robust in handling pose and illumination changes than either one of them by itself. We show how the algorithm performs on a real surveillance scenario where the face undergoes various pose and illumination changes. The algorithm runs in real-time at 20 fps on a standard 3.0 GHz desktop PC.

Analog dial gauge reader for handheld devices

In a typical process industry, there are several critical dial gauges wherein the readings are to be monitored on a periodic basis. Most of these dial gauges are still analog in nature and currently these are being monitored manually. The manual capture of readings is a tedious task and is also prone to error. This paper outlines an algorithm that can automatically identify the dial gauge readings using image captured by a PDA device. This will help in reducing the error in readings as well to provide a record of the actual reading for future reference. Our algorithm uses polar representation of the dial gauge image to identify the needle position as well as the start position of the dial gauge. Once these are obtained the reading of the dial gauge can be estimated with prior calibration information. The algorithm was tested on multiple dial gauge images captured in a local chiller plant and is shown to obtain the readings reliably in about 95% of these images.

Flare monitoring for petroleum refineries

In petroleum refineries, it is a common practice to flare up the exhaust gases before releasing them to the atmosphere in order to reduce the environment pollution. The area or volume of the flare indicates the quantity of gas that is getting released in the refining process and the color of the flare at any given time is decided by the constituents of the gases in the flare and the volume of the flare indicates the quantity of gas that is released. In an indirect way, these parameters of flare indicate the performance of refining process. Presently, the flare is manually observed by the operator and doing so reliably on a 24×7 basis is a difficult job. In this paper we propose an algorithm1 to automate this effort using video analytics and provide the alarms in case of any abnormal flaring event. The flare detection is done by estimating models for foreground and background regions and the color of the flare is analyzed by using classifiers with kernel function, such as support vector machine (SVM). The performance of these algorithms is tested on various data sets collected from refineries.