We consider in this paper the improvement of side-attack mine detection by performing confidence level fusion with data collected from vehicle-mounted forward-looking IR and GPR (FL-GPR) sensors. The mine detection system is vehicle based, and has both IR and FL-GPR sensors mounted on the top of the vehicle. The IR images and FL-GPR data are captured as the vehicle moves forward. The detections from IR images are obtained from the Scale-Invariant Feature Transform (SIFT) and Morphological Shared-Weight Neural Networks (MSNN) depending on target characteristics, and those from FL-GPR are derived from the FL-GPR SAR images through object-tracking.
Since the IR and FL-GPR alarms do not occur at the same location, the fusion process begins with each IR alarm and looks at the nearby FL-GPR alarms with confidences weighted by values that are inversely proportional to their distances to the IR alarm. The FL-GPR alarm with the highest weighted confidence is selected and combined with the IR confidence through geometric mean. An experimental dataset collected from a government test site is used for performance evaluation. At the highest Pd and comparing with IR only, fusing IR and FL-GPR yields a reduction of FAR by 26%. When the Hough transform is applied to reject the IR alarms that have irregular shapes, the fusion results provides a reduction of FAR by 35% at the highest Pd.
There has been significant amount of study on the use of Ground-Penetrating Radar (GPR) for forward-looking landmine detection. This paper presents our analysis of GPR data collected at a U.S. Army site using the radar system developed by Planning Systems Inc. (PSI). One property of forward-looking systems is that a target appears in multiple radar images at different distances. To exploit this property, we divide the distance range in the radar images into a number of distance bands. Identification of potential targets is first performed in each distance band independently. Our algorithm then tracks these potential targets through multiple distance bands and computes weighted averages of their geometrical features. The persistence property of the targets is used to further reduce false alarm rates by removing potential targets that only appear spuriously. Results of landmine detection, including performance on blind test lanes, are presented.
There has been significant amount of study on the use of Ground-Penetrating Radar (GPR) for forward-looking landmine detection. This paper presents our analysis of GPR data collected at a U.S. Army site using the Synthetic Aperture Radar system developed by Stanford Research Institute (SRI). Various types of features are extracted from the GPR data and investigated for their abilities to distinguish buried landmines and false alarms; the list include intensity and local-contrast features, fuzzy geometrical image features, ratio between co-polarization and cross-polarization signals, and features obtained using two different approaches of polarimetric decomposition. We also describe the feature selection procedures employed to find subsets of features that improve detection performance when combined. In addition, our analysis indicates that images formed with different frequency bands have different qualities, and that the selection of proper frequency bands can significantly improve the detection performance. Results of landmine detection, including performance on blind test lanes, are presented.
Detection of tripwires is an active area of investigation. Researchers at the University of Missouri and the University of
Florida are jointly pursuing numerous approaches to detect both the trip wires and the mines to which they are connected. In this paper, we discuss issues related to the detection of tripwires both on a frame-by frame basis and within image sequences. A large data collection was performed under various environmental conditions. Using
algorithms that operate on visible and near infrared imagery, several metrics are explored to categorize tripwire detection performance. Currently, the detection algorithm utilizes the Hough transform to detect line-like structures and then scores these candidates to distinguish between wires and linear background objects. Adaptations to the Hough transform are discussed to add robustness and to decrease the computational load. Within the sequence analysis, emphasis is placed on the use of fuzzy logic rules to integrate over time. Results of several experiments in the outdoor settings are
described and analyzed.
The utility of acoustic-to-seismic coupling systems for landmine detection has been clearly established. In this approach, laser Doppler vibrometers (LDV) are used to measure the different responses to acoustic excitation in ground regions with and without buried landmines. Currently, for most applications, only the magnitude of the surface velocity is investigated and used to construct recognition algorithms. Recently, we introduced phase-based features in the classification scheme, significantly lowering false alarm rates at given detection probabilities. In this paper, we present
modeling equations that explain the phase features for ground areas both with and without buried landmines from the perspective of harmonic oscillator models. We also describe the image processing techniques applied to velocity data collected in the time domain with a moving LDV array. The observed signatures are also compared with the prediction of the models described. We also construct classifiers with only magnitude information and both magnitude and phase information for this time-domain data set. Classification results indicate that we can combine magnitude and phase features to improve the detection of buried mines while reducing false alarms. We also find that using phase information improves the distinction between ground regions with buried landmines or man-made clutter objects.
Acoustic-Seismic methods for landmine detection are under intensive investigation. Data collected by the University of Mississippi have by processed by a variety of investigators with excellent results in many cases. Increases in performance are sought based on an understanding of the physical principles leading to the differences between the vibrational velocities of soil over buried landmines and over locations not above landmines. Donskoy suggested modeling the physical system using damped harmonic oscillators. This model suggests a use of magnitude and phase information in image processing algorithms for detecting. In this paper, some methods for incorporating magnitude and phase into image processing algorithms are described and demonstrated. Previous algorithms relied on magnitude only. Increased performance is achieved by incorporating phase into the algorithms.
A variety of sensors have been investigated for the purpose of detecting buried landmines in outdoor environments. Mines with little or no metal are very difficult to detect with traditional mine detection systems. Ground Penetrating Radar (GPR) sensors have shown great promise in detecting low metal mines and can easily detect metal mines. Unfortunately, it can still be difficult to detect low-metal mines with GPR due to very low contrast between the mine and the surrounding medium. Acoustic-seismic systems were proposed by Sabatier et.al. and have also shown great promise in detecting low metal mines. There are now a wealth of references that discuss these systems and algorithms for processing data from these systems. Therefore, they will not be discussed in detail here. In fact, low-metal mines are easier to detect than metal mines with this acoustic-seismic systems. Low metal mines that are difficult for a GPR to detect can be quite easy to detect with acoustic-seismic systems. Sensor fusion with these sensors is of interest since together they can find a broader range of mines with relative ease. The algorithmic challenge is to determine a strategy for combining the multi-sensor information in a way that can increase the probability of detection without increasing the false alarm rate significantly. In this paper, we investigate fusion of information obtained from GPR and acoustic-seismic on real data measured from a mine lane containing three types of buried landmines and also areas containing no landmines. Algorithms are applied to data acquired from each sensor and confidence values are assigned to each location at which a measurement is made by each sensor. The GPR is used as a primary sensor. At each location at which the GPR algorithm declares an alarm, a modified likelihood-based approach is used to increase the GPR derived confidence if the likelihood that a mine is present, defined by the acoustic-based confidence, is larger than the likelihood that no mine is present. If the acoustic-derived confidence is very high, then a declaration is made even if there is no GPR declaration. The experiments were conducted using data acquired from the sensors at different times. The acoustic-seismic system collected data over a subset of the region at which the GPR collected data. Results are given only over those regions for which both sensors collected data.
Detection of tripwires is an active area of investigation. Researchers at the University of Missouri and the University of Florida are jointly pursuing numerous approaches to detect both the trip wires and the mines to which they are connected. Utilization of robotic vehicles capable of performing this task is one of the goals of this project. In this paper, we discuss issues related to the embedding of current versions of our tripwire detection algorithms into a small and inexpensive robot testbed for real-time experimentation. The robot is based a simple remote-controlled truck where the remote control unit has been replaced by a standard microcontroller. Sensors are added to assist navigation tasks, handled by the microcontroller, and the tripwire detection algorithms are implemented on a laptop PC with video input. There are several sophisticated algorithms that are being investigated for robust tripwire detection. The current detection algorithm that has been pruned down to run in real-time on the robotic platform consists of a Hough transform to find candidate lines followed by post-processing to score the candidate lines for the likelihood that they correspond to a trip wire. Upon detection, the robot is given a command to stop. Results of several experiments both indoors and outside in a variety of settings are described and analyzed.