Abstract

This paper explores a tracking method of broccoli heads that combine a Particle Filter and 3D features detectors to track multiple crops in a sequence of 3D data frames. The tracking accuracy is verified based on a data association method that matches detections with tracks over each frame. The particle filter incorporates a simple motion model to produce the posterior particle distribution, and a similarity model as probability function to measure the tracking accuracy. The method is tested with datasets of two broccoli varieties collected in planted fields from two different countries. Our evaluation shows the tracking method reduces the number of false negatives produced by the detectors on their own. In addition, the method accurately detects and tracks the 3D locations of broccoli heads relative to the vehicle at high frame rates