A Binary Local Feature Descriptor Using Convex Optimization And Its Application To Real-time Crop Row Detection

Coding a local patch of an image into a numeric sequence, local feature descriptor is a primary feature in image processing techniques, which has been widely applied to many applications, such as image segmentation and 3D reconstruction. With the rapid development of mobile devices, embedded in which the computer vision techniques are constrained with memory capacity and computational speed. Under this circumstance, binary local feature descriptors have attracted more and more attentions, thanks to their low memory requirement but fast matching speed. However, binary local feature descriptors suffer from the problems with low accuracy and low robustness.We propose a novel algorithm in this dissertation, in which th e candidate pooling regions of binary local feature descriptors are determined via convex optimization. Accordingly, the final binary local feature descriptors are selected by ranking the performance of the above candidates on the training data in terms of high discriminative capability but low correlation. The experimental results on benchmark datasets have shown that our proposed binary local feature descriptors outperform the representative methods.Our research has already been brought into practice – real-time crop row detection in agricultural machinery. Intelligent agricultural machinery not only relieves intensive labor for humans but also reduces damages to crops. In this application, the baseline of the crops row is localized through analyzing histogram projection of corresponding binary images. The boundaries of the crops are then determined by analyzing the pixel values of the lines parallel to the baseline. To speed up the detection procedure in machine cultivation, our proposed binary local feature descriptor is employed to match a pair of corresponding patches between consecutive frames, such that the row detection can be analyzed within a local area rather than the whole image. The field assessments on site have verified that our method is capable of meeting the requirements for practical use.