[P]DATA FUSION FOR LONG RANGE TARGET ACQUISITION

Đây là một công trình nghiên cứu về khoa học quân sự. Tôi sẽ lần lượt đăng tải toàn bộ công trình, và sau đó dịch sơ qua để các bạn cùng tham khảo. Bạn nào dịch hộ nữa thì tốt quá

Nguồn : http://www.rma.ac.be/TELE/staff/patrick.verlinde.html

DATA FUSION FOR LONG RANGE TARGET ACQUISITION
Patrick Verlinde, Dirk Borghys, Christiaan Perneel, Marc Acheroy
Signal and Image Centre
Royal Military Academy
Renaissancelaan 30,.
B1000 Brussels
Belgium

​

SUMMARY

An approach to the long range automatic detection of vehicles, using multi-sensor image sequences is presented. The algorithm we use was tested on a database of six sequences, acquired under diverse operational con***ions. The vehicles in the sequences can be either moving or stationary. The sensors are stationary, but can perform a pan/tilt operation. The presented paradigm uses data fusion methods at four different levels (feature level, sensor level, temporal level and decision level) and consists of two parts.

The first part detects targets in individual images using a semisupervised approach. For each type of sensor a training image is chosen. On this training image the target position is indicated. Textural features are calculated at each pixel of this image. Feature level fusion is used to combine the different features in order to find an optimal discrimination between target and nontarget pixels for this training image. Because the features are closely linked to the physical properties of the sensors, the same combination of features also gives good results on the test images, which are formed of the remainder of the database sequences. By applying feature level fusion, a new image is created in which the local maxima correspond to probable target positions. These images coming from the different sensors are then combined in a multi-sensor image using sensor fusion. The local maxima in this multi-sensor image are detected using morphological operators. Any available prior knowledge about possible target size and aspect ratio is incorporated using a region growing procedure around the local maxima. A variation to this approach, that will also be developed in this paper, combines the previous feature and sensor level fusion, by extracting the features in each sensor as before but using the feature level fusion directly on the combination of all features from all sensors in what is sometimes called a « super feature
vector ». Tracking is used in both cases to reduce the false alarm rate.

The second part of the algorithm detects moving targets. First any motion of the sensor itself needs to be detected. This detection is based on a comparison between the spatial cooccurrence matrix within one single image and the temporal cooccurrence matrix between successive images in a sequence. If sensor motion is detected it is estimated using a correlationbased technique. This motion estimate is used to warp past images onto the current one. Temporal fusion is used to detect moving targets in the new sub-sequence of warped images. Temporal and spatial consistency are used to reduce the false alarm rate.

For each sensor, the two parts of the algorithm each behave as an expert, indicating the possible presence of a target. The final result is obtained by using decision fusion methods in order to combine the decisions of the different experts. Several « k out of
n » decision fusion methods are compared and the results evaluated on the basis of the 6 multi-sensor sequences.



Được RandomWalker sửa chữa / chuyển vào 00:11 ngày 30/08/2003

1 INTRODUCTION
Long range automatic detection of vehicles is of great military importance to modern armed forces. The most critical factor of any system for automatic detection is its ability to find an acceptable compromise between the probability of detection (= 1 - probability of a miss) and the number of false alarms. This is the classical trade-off one finds in binary hypothesis testing between the two types of error one can make : the false rejection (FR : which corresponds here to a miss : there is a target, but it has not been found) and the false acceptance (FA : which is in this case the same as a false alarm : there is no target, but the system thinks there is one). In a single sensor detection system it is well known that if one reduces one type of error, the other type of error automatically increases. A possible way-out of this deadlock is to use more than one sensor and to combine the information coming from these different ô experts ằ. This combination or (data) fusion can be done on different levels. In this paper, only the (common) case of a centralised fusion processor with all its sensors connected in parallel will be considered.
Được RandomWalker sửa chữa / chuyển vào 00:24 ngày 30/08/2003

In the specific data fusion literature [1-5] one often distinguishes between the following (or equivalent) fusion levels : low level fusion (also called score or measurement level fusion), medium level fusion (which includes feature and sensor level fusion), high level fusion (also called decision level fusion) and temporal level fusion. As can be expected, in real (-time) applications, there is a trade-off to be made between the amount of information that can be combined and the bandwidth necessary to communicate all this information to the centralised fusion processor. The lower the level of fusion, the more information is available to be combined, but the larger becomes the bandwidth necessary to communicate with the centralised fusion processor (or for a fixed bandwidth, the slower becomes the fusion process). Vice versa one sees that when the level of fusion gets higher, the available information diminishes, but so does the necessary bandwidth. Furthermore not all data fusion levels are always applicable. For instance, if low level fusion is going to be used, care must be taken to combine only similar entities (scores, measurement results,...). It is therefore impossible to use low level fusion to combine the raw results coming from two (or more) totally different sensors (e.g. an imaging sensor and a range finder). But this constraint doesnõ?Tt exist any longer on the decision level, where each sensor is considered as a separate ô expert ằ, who decides on his own. In the special case of target detection where the ô hard ằ binary decision rule is used (the ô hard ằ decision is indeed bi-valued : target present (1) or not (0)), the central fusion processor contents itself to combine only the decisions (the 1õ?Ts and the 0õ?Ts) coming from different sensors, without considering the type of sensor. As a general conclusion concerning the different data fusion levels, one can state that all different fusion levels have their importance and their specific applicability domain.
Based on these considerations, we have tried to use data fusion on several levels to try to optimise the use of the available data. That is basically why this paper describes an approach to tackle the previously exposed problem using four different data fusion techniques related to several levels : feature level fusion, sensor level fusion, temporal level fusion and decision level fusion. The only fusion level that is not used in this paper is the low level fusion. This technique (in the form of pixel level fusion) is mainly used in remote sensing applications [6, 7]. In the main approach, we do however use two different medium level fusion techniques. In the following sections the use of these different data fusion techniques will be explained in more detail.
Được RandomWalker sửa chữa / chuyển vào 00:27 ngày 30/08/2003