Crop-row detection is a central element of weed detection and agricultural image processing tasks. With the increased availability of high-resolution imagery, a precise locating of crop rows is becoming practical in the sense that the necessary data are commonly available. However, conventional image processing techniques often fail to scale up to the data volumes and processing time expectations. We present an approach that computes regression lines over large windows that surround pixels. Pixels for which the excess green value exceeds a threshold are viewed as being part of plants. Those pixels are then processed as if they were data points in a statistical analysis. Note that the image may have to be rotated by 90 degrees to make this approach useful. Once the regression parameters of slope and offset have been computed, the central positions of the crop rows is inferred. Note that the resolution of the imagery is preserved in the process, because sliding window techniques use every possible sub-window of the image, i.e., there is one window associated with every pixel in the image. Our approach is highly computationally efficient due to an aggregation strategy that scales logarithmically in the window size. Aggregates of spatial coordinates x, x2, y, and xy are initially computed over sliding windows of size 2x2, and then the resulting windows are aggregated iteratively to generate all possible sub-windows of size 4x4, 8x8, 16x16 etc. In our example analysis, 64x64-sized windows capture individual crop rows without overlapping with multiple rows. Computing aggregates for windows of size 64x64 only requires 6 basic passes of the data.  By working on memory-resident slices of the imagery we further avoid the computational cost that would result from multiple disk accesses. At no point does our approach rely on typical image processing steps such as those involved in image segmentation. That means that the algorithm is intrinsically designed to allow being deployed in a real-time processing context.