Options
Masturah Tunnur Mohamad Talib
Human localisation in an outdoor environment using radio frequency tomography
2022
,
Masturah Tunnur Mohamad Talib
Localisation of humans in an outdoor environment can offer great capabilities especially in security and perimeter surveillance applications. However, localizing humans in a nonlinear environment is a challenge. Although, GPS and CCTV successfully detect human location, these two techniques provide low localisation accuracy due to signal loss and camera view limitations. Device-free localisation (DFL) technique has been introduced to overcome these problems. This is because the DFL approach has the capability to detect the human body wirelessly in all environmental conditions and there is no losing signals problem as faced by GPS. Despite that, the accuracy of the DFL approach is still low due to the problem of variation in radio signal strength indicator. To overcome this problem, Radio Tomographic Imaging has been introduced. Basically, this approach characterized the differences in radio frequency (RF) response by exploiting the radio frequency attenuation. The differences in RF response occur due to the signal obstructed by the human body. Conventionally, RTI uses the Linear Back Projection algorithm (LBP) to reconstruct the tomographic image. However, it produces a low-quality image due to the ill-posed inverse problem caused by back-projection and the smearing effect due to the overlapping image. Several regularization approaches have been introduced by other researchers to improve the quality of tomographic images. These regularization techniques have been used to eliminate the smearing effect on the RTI image. However, the resulting image is still blurred because the target occupies only a small amount of space compared to the entire area monitored. Therefore, the new image reconstruction algorithms called as Hybrid Radio Tomographic Imaging Technique (HRTI) and Modified Hybrid Radio Tomographic Imaging Technique (HRTI-M) have been proposed to solve the RTI problem. Through these techniques, the area error analysis score for HRTI is lower than 3% and lower than 1% for HRTI-M. The benefit of using the proposed methods is the point location of the human can be identified from the reconstructed image and this will contribute to increasing the localisation accuracy. These RTI data are introduced to the Deep Neural Network (DNN) classification approach to improving localisation accuracy. This approach is known as the RTI-DNN approach. In this approach, the HRTI-M is used to eliminate the variation of sensor data and improve the quality image of RTI. While, the DNN is used to reduce the classification error due to variation in sensor data and increase the accuracy of localisation. This RTI DNN was then compared with RTI based on the Artificial Neural Network (RTI-ANN) approach to evaluate the performance of the proposed method. From the classification results, it is found that the performance of RTI-DNN is better than RTI-ANN which is 88% compared to RTI-ANN with only 64% accuracy.