This Case Study is made by the following student of VIT, PUNE

Name - Snehal Badhan

Roll no. - 14

Introduction:

Signal and image processing plays an important role in every field. Signal processing basically deals with the operations on performed on signals. We can extract the data from what we want from any signal in the form of image using various image processing techniques.

When we talk of target detection and tracking it is very important in every field and one of the most important is in the military where many threats are present and without risking one’s life target is to be detected. There are various techniques and algorithms used for target detection and tracking and how signals are received and the targets are detected. Some of them are discussed below.

CASE STUDY TITLE 1: SIGNAL PROCESSING TO IMPROVE TARGET DETECTION USING GROUND-PENETRATING RADAR (GPR)

This case study basically focuses on improving target detection using GPR pre-processing tasks.

What is GPR?

Ground penetrating radar (GPR) is a non-intrusive technique used to obtain information about the data below the surface of the earth. The technique GPR relies upon the transmission of electromagnetic energy into the earth. The fields such as archaeology, geology, civil engineering, and military applications uses GPR technique for target detection. Detection of buried landmines, road layer thickness measurement, depth of underground pipes, and detection of chemical spills are some of the interesting applications where GPR technique is been used.

Also, one of the limitation of GPR technique is that it can be very difficult for a non-expert user to extract information from the raw data.

Block Diagram of GPR

The GPR technique consists of five main components: transmitter/receiver radar electronics, transmitter and receiver antenna assembly, timing module, analogue sampling circuitry, and data acquisition. Below shown is the block diagram of GPR:

Antenna Hardware and Operation

The GPR transmitter electronics produces a 1-2ns pulse, which is transmitted and received via a pair of shielded bistatic bow-tie antennas. In this case study the centre frequency of the pulse, as determined by the dimensions of the antenna, is 800 MHz with a bandwidth suitable for propagating through coal.

Equivalent Time Sampling

An equivalent time sampling technique is employed which uses analogue electronics to acquire the received signal at a rate significantly lower than direct sampling. An equivalent sampling frequency is necessary so that the distance to a target can be estimated.

Data Acquisition

A falling edge synchronisation pulse is generated by the analogue sampling electronics at the start of a new received waveform. The synchronisation pulse is used by the acquisition software to form consecutive data realisations. The data can then be ported into other signal processing packages for further processing and analysis as necessary.

Data Processing in GPR

Several pre-processing techniques are described as follows. These are important to simplify the interpretation of GPR data. These basically include jitter correction, averaging (also referred to as stacking), background signal removal, and time-varying gain.

Jitter Correction:

For the alignment of both the received and the background signals Jitter Correction is implemented.

Averaging:

This technique is used to reduce the noise bandwidth of the received signal.

Background Removal:

The background signal can be considered as the calibrated signal which can be observed even when the target is not present.

The background signal is then subtracted from the received signal and thus resulting in a wavelet shape signal with time delay proportional to the distance of the target.

Time-Varying Gain:

Sometimes, there is an effect of reducing the amplitude of pulses which is caused due to targets are far away from the antennas. For the best results and compensation of this effect, a non-linear gain is applied to the received waveform.

The time-varying gain function requires the addition of two components, an exponential and a 4th power. The exponential component is designed to maintain the close-range signal, while the other is to amplify the attenuated signal at distances furthest from the antennas.

Matched Filter Based Detector:

The filter is used to detect when the signal component of the received waveform is correlated with a reversed time shift of the transmitted signal. The point of greatest correlation is taken as the reflection from the target.

Transmitted Pulse Identification

Conclusion

The use of radar pre-processing has to lead to an improvement in distance estimation even before formal processing has begun.

The limitation of the conventional matched filtering technique applied here is that only one target can be detected as the point of the highest correlation is taken as the reflection from the target.

GPR can also be used for:

Detecting voids and concrete homogeneity.

Helps in detecting environmental and natural structures such as sinkholes, soil structures, water tables, saltwater infiltration, and groundwater channels.

Measuring ice thickness which is important in areas where ice roads exist for part of the year and arctic oil exploration.

Detecting underground storage tanks

Using GPR has many advantages including:

· No health hazards

· Instant results

· Provides depth estimates

· Can see through up to 2-feet of concrete

· Can see pipes below the floor

CASE STUDY TITLE 2: ONE-DIMENSIONAL MORPHOLOGY-BASED METHOD FOR SELECTIVE TARGET DETECTION

Typical images contain small targets surrounded by irregular backgrounds in a large area, such as stars (target) with a nebula or a galaxy (interference or background), as shown in (Figure a). A point spread function (PSF), expressed by Equation (3), can be employed to approximate a small target, which is basically a connected area with a decrease in brightness from the target centre to its edge, as expressed by a Gaussian distribution

Image properties in the spatial domain and the frequency domain: (a) original image with stars of different sizes, as well as a nebula and a galaxy; FFT results of (b1) the original image, (b2) single-pixel noise, (b3) background noise, and (b4) targets. Centers and edges of the FFT results correspond to lower and higher frequencies, respectively. Illustration of RbR image processing for targets in row i with different sizes: (c1) extraction of a target with a larger size and (c2) extraction of a target with a smaller size. Insets are the zoom-in view of the extracted targets in a.

where I₀ is the total energy pertaining to the target, (x₀, y₀) is the expected center of the target and σ is the Gaussian radius, which is related to the concentration of the PSF energy distribution.

There are different types of noise and interference within an image containing the desired small target, and their spatial scales are chosen as the criterion for classification. The image background, large scale noise such as fixed pattern noise comparable to the whole image in spatial scale, and large-scale interferences are all treated as background noise. Figure 2b explicitly shows the composition of an image with N rows and N columns in the frequency domain. The center and edge correspond to the lower frequency and higher frequency, respectively, and the intensity is the logarithmic value of the fast Fourier transform (FFT). When performing the FFT on a 2D image, the 1D FFT of each row is taken to form an intermediate image, followed by the 1D FFT of each column in the intermediate image. the desired target signal is overwhelmed by the background noise, whose FFT result, shown in Figure 2b3, features dominant low-frequency components due to the continuous distribution of the background noise signal. After filtering all undesired parts from the original image, the FFT result of the desired targets with size K is plotted in Figure 2b2 with small components ranging from low frequency to intermediate frequency.

Image processing was achieved by a mathematical morphology-based image processing approach. Erosion and dilation are two basic operations in mathematical morphology.

where f(x, y) and h(x, y) are set to represent the gray value (energy intensity) of images f and h at a certain point (x, y), respectively. b is the SE, D_b is the domain of b, and D_f is the domain of f.

Erosion can be used to remove the single-pixel noise or smaller objects in the image, while through a combination of erosion and dilation, background noise can also be obtained for target enhancement.

CASE STUDY TITLE 3: TARGET DETECTION AND TRACKING OF MOVING OBJECTS FOR CHARACTERIZING LANDSLIDE DISPLACEMENTS FROM TIME-LAPSE TERRESTRIAL OPTICAL IMAGES

TDT methods are used to estimate the displacements of discrete features over time, either natural (rock blocks, large fractures) or man-made (benchmarks). The selected objects should stand out by their radiometric properties and ensure their effective and accurate detection. The main advantage of TDT methods over image correlation techniques is the ability to measure complex displacement patterns over space and time for large deformation, high signal-to-noise ratios, changes in the target's shape and the lag of images over time for low view angles.

Algorithm of TDT Method

Step 1: Image registration

Accurate image co-registration is a key element in the application of change detection methods because it determines the reliability of the displacement estimates. Image co-registration generally consists of correcting the camera motion induced by wind or temperature variations and transferring each image of the time series into the geometry of the reference image.

Image registration basically is performed using four methods:

1) Detection of time invariant points in the image: e Harris corner detector algorithm is used because it is computationally more efficient for studies that use fixed cameras. It is based on the first order image derivatives in the x–y directions in gray-scale images. A cross-correlation in the Fourier domain is then performed on the Harris corner point coordinates to attain a matching resolution of 0.1 pixel.

2) Feature matching: This stage consists of establishing the correspondence between the invariant points on the complete image time series. Commonly matched points are then correlated and points that are strongly correlated in both x–y directions are paired together.

3) Estimation of the mapping model: Estimates the best fit of the harris corner point pairs in the image time series. The most appropriate model is chosen by comparing different (affine, projective, polynomial non-reflective, reflective similarity)

4) Projection of the object coordinates: This approach avoids errors due to image interpolation and resampling during the projection, preserves the radiometry of the image and reduces the processing time

Step 2: Automated feature detection and tracking

The larger the size, the greater the probability that the object can be detected in the search window but also that it can be mixed with other objects. Then a background image is then constructed by applying a rank-order filter in which each pixel is replaced with its darkest neighboring pixel. This mathematical morphological operation, called “erosion”, only retains the large trend of the initial image. A new thumbnail is then created by subtracting the eroded image from the initial image. Radiometric stretching is applied to produce grayscale intensities ranging between 0 and 255. This operation, called “opening”, enhances the contrast, allows better object detection by thresholding, and compensates for changes in the illumination conditions. A binary image is created by applying the Otsu method, which minimizes the grayscale variance inside the two classes assumed to make up the image.

Step 3: 3D photogrammetric restitution

External orientations of the cameras are then computed. In the case of stereo- (or multiple) views, the 3D restitution is performed by stereoscopy. With the camera orientation information, it is then possible to determine the local coordinates (X,Y,Z) of an object from two pairs of coordinates.

Conclusion:

Target Detection and Tracking (TDT) method that is based on simple binary image processing and is designed as a complementary technique to image correlation. This method allows tracking natural or man-made targets in a time series of images.

the TDT approach does not provide spatially continuous information, it provides 1) a quantification of the object displacements at the same order of precision as image correlation (sub-pixel accuracy) and 2) information in regions where image correlation fails because of too large ground displacements. A sensitivity analysis reveals that the major sources of uncertainty are camera movement and/or lens distortion and not the TDT method itself.

References:

1. https://www.researchgate.net/publication/27464986_Signal_Processing_to_Improve_Target_Detection_Using_Ground_Penetrating_Radar

2. https://www.nature.com/articles/lsa20186

3. https://www.sciencedirect.com/science/article/abs/pii/S001379521400009X

4. https://www.researchgate.net/publication/283517014_Target_Detection_Using_Image_Processing_Techniques#:~:text=The%20target%20is%20detected%20using,lighting%2C%20shadows%2C%20and%20distance.

5. https://www.hindawi.com/journals/js/2017/8231314/

6. https://pubmed.ncbi.nlm.nih.gov/33203032/