Improved image processing-based crop detection using Kalman filtering and the Hungarian algorithm

There is increasing interest in the use of image processing techniques for crop detection in intelligent weeding applications. An effective system for crop detection requires a high degree of adaptability to challenging circumstances such as different weather conditions and image capture conditions (vibration, variations in speed, etc.). To achieve the goal of a robust crop detection system, we have extended a previously-developed detection algorithm that is based on a combination of color-space and shape analysis, through the addition of object tracking. While the previous algorithm performed well in general, performance in sunny conditions was not as robust, opening up the possibility of improvement. The tracking algorithm consists of two steps. Firstly, we apply Kalman filtering to predict the new position of an object (a cauliflower plant in this case) in video sequences. Secondly, we use a data association algorithm (the Hungarian algorithm) to assign each detected crop that appears in each image to the correct crop trajectory. The recall matrix was used to evaluate the detection and tracking performance. With the help of tracking algorithm, detection failures were reduced, especially in sunny conditions, such that overall detection performance was raised from 97.28 to 99.3404%.