Constant False Alarm Rate (CFAR):

Monday, March 18, 2024 12:41 AM

Paper title: Detecting Cars in VHR SAR Images via Semantic CFAR Algorithm What is CFAR algorithm The OTSU algorithm is an image thresholding algorithm proposed by Nobuyuki Otsu. It is an
unsupervised learning method that is widely used in image processing. The algorithm selects
Summary: • CFAR stands for Constant False Alarm Rate. thresholds by maximizing the ratio of between-class variance and within-class variance. It is
• The paper proposes a semantic CFAR algorithm for car detection in VHR SAR images, which improves the detection result by • It is a detection algorithm commonly used in radar signal processing to particularly effective in adapting to different images and has good performance in image
using the spatial semantic relationship between the strong scattering surface and shadow of the same target. detect targets in the presence of noise and clutter. segmentation tasks. In the context of car detection in VHR SAR images, the OTSU algorithm is
• CFAR algorithms estimate the statistical properties of the background employed to search for dark pixels, which correspond to the shadow regions of cars. By using
• The algorithm removes isolated bright points and false alarm targets by analyzing the sliding reference window of the strong clutter and determine a threshold for target detection based on these the OTSU algorithm, the dark pixels in the SAR image can be effectively identified and
scattering surface and matching spatial semantic relationships. statistics. segmented, allowing for the detection of the shadow areas associated with cars
• The CFAR algorithm is widely used in SAR (Synthetic Aperture Radar) image
• The proposed semantic CFAR algorithm demonstrates lower false alarms compared to the traditional CFAR algorithm for car processing for target detection.
detection in VHR SAR images. • In the context of car detection in VHR SAR images, the CFAR algorithm is
employed to search for bright pixels corresponding to the strong scattering DATASET:
surface of the car [1]. https://www.sandia.gov/radar/pathfinder-radar-isr-and-synthetic-aperture-radar-sar-
• The paper highlights the key characteristics of cars in SAR, such as strong scattering (layover or double-bounce) and shadow systems/complex-data/
effects caused by the side-looking property of SAR sensors. • The CFAR algorithm is used in combination with the OTSU algorithm,
which is an unsupervised learning method for image thresholding, to
• The paper also mentions that car detection techniques based on optical imagery are vulnerable to weather conditions, while SAR search for dark pixels corresponding to the shadow of the car [1].
techniques can be applied in all weather conditions. • The proposed semantic CFAR algorithm for car detection in VHR SAR NOTES:
images improves the detection result by considering both the strong The images used in the experiments are described as the Tijeras Arroyo Golf Course
• The proposed semantic CFAR method outperforms the CFAR algorithm in terms of false alarm rate, as shown in experiments on a scattering features and the shadow features of the target [2]. and the Wyoming Gated Entrance, KAFB, both having a resolution of 1638 x 2510
Mini SAR image • The semantic CFAR algorithm reduces false alarm targets by establishing pixels
the semantic relationship between the strong scattering features and the
Modelling the car in VHR: shadow features
>
• The key characteristics of cars in SAR are the strong scattering (layover or double-bounce) and shadow effects caused by the side-
looking property of SAR sensors .
• A simplified car model is shown in Figure 2, where different regions of the car are labeled, including the strong scattering areas
from the windscreen and head, and the shadow area behind the car.
• The elevated car occludes parts of the surface behind it from the radar beam, resulting in the shadow area .
• The length of the tail of the car determines the extent of the strong scattering area .
• The paper mentions that the key characteristics of cars in SAR are similar to buildings, which have been described in the
literature.
• The modelling section provides a foundation for understanding the spatial semantic relationship between the strong scattering
surface and shadow of the car, which is utilized in the proposed semantic CFAR algorithm for car detection

Semantic Primitives in the SAR Image:

• The SAR image of a car consists of two semantic primitives, namely the strong scattering surface and the shadow.
• The strong scattering surface is composed of bright pixels in the SAR image, which represent areas of high radar reflectivity.
• On the other hand, the shadow is made up of dark pixels in the SAR image, indicating areas of low radar reflectivity.
Detection of Bright Pixels using CFAR
• To identify the bright pixels corresponding to the strong scattering surface, a detection technique called Constant False Alarm
Rate (CFAR) is employed.
• CFAR is a statistical method that compares the radar return from a target with the surrounding clutter to determine if the target
is present.
• In this case, CFAR is used to search for bright pixels in the SAR image, which are indicative of the strong scattering surface of the
car.
Detection of Dark Pixels using OTSU Algorithm
• The dark pixels in the SAR image, which form the shadow of the car, are identified using the OTSU algorithm.
• The OTSU algorithm is an image thresholding technique proposed by Nobuyuki Otsu.
• It selects the optimal threshold value by maximizing the ratio of between-class variance and within-class variance.
• In simpler terms, the algorithm finds the threshold that best separates the dark pixels (shadow) from the rest of the image based
on their intensity values.
• By applying the OTSU algorithm, the dark pixels in the SAR image can be accurately detected, allowing for the identification of the
shadow region of the car.
Importance of Bright and Dark Pixel Detection
• The detection of bright pixels (strong scattering surface) and dark pixels (shadow) in the SAR image is crucial for various
applications, such as target recognition and object detection.
• By identifying the strong scattering surface, which represents the main body of the car, it becomes possible to distinguish the car
from other objects in the SAR image.
• Similarly, the detection of the shadow region provides additional information about the car's shape and orientation, aiding in its
accurate identification and analysis. Fig. 2

Review 2 Page 1