However, with a large
threshold, the missed detection rate will also increase. Similarly, having a large
probing area can also improve the performance since the statistical analysis
can be more accurate if more values of event predicates can be collected.
But the communication overhead for a large probing area will also increase
significantly
Thus, an interesting topic here would be the selection of the threshold
value t0 and the size of probing radius R. To reach a satisfactory operation,
one can pre-calculate the performance curves and store the selected threshold
and probing radius into sensors such that the statistical approach can satisfy
the preferred performance criteria.
152
Chapter 6 Boundary Detection for Sensor Networks
2.2 Image processing based approach
A number of edge detection schemes have been designed and analyzed in the
image processing literature. A typical method is to make use of a high pass
filter (e.g., Prewitt, Sobel filters) to detect edges in an image. By convoluting
an image of interest with a high pass filter, regions of high spatial gradient
that correspond to edges will be emphasized.
Usually, a high pass filter G(x, y) consists of at least a pair of 3x3 convolution
masks, one mask for each of the two orthogonal axes. For example,
the Prewitt edge detector defines mask Gx to calculate the gradient along the
horizontal orientation, and Gy is used to estimate the gradient of the vertical
orientation.
Pages:
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268