Large Lump Detection Using a Particle Filter of Hybrid State Variable

Abstract

This paper presents a particle filter based solution to the problem of detecting large frozen lumps in an image sequence, taken of the feed to a crusher, which is used for size reduction of oilsand ore. In this application, the objects of interest, i.e., large frozen lumps, are characterized by a high level of image noise, irregular shapes, and uneven and variable surface texture. In addition, more than one large lump can be present in the scene. Our proposed solution integrates evidence of the presence of large lumps over time, by adapting an existing Bayesian framework for joint object detection and tracking. To implement the particle filter, we formulate an application-specific observation model that is required by the Bayesian tracker. Our experimental results show that the proposed solution is capable of detecting multiple large lumps reliably, and that it has the potential of preventing the oilsand crusher from being jammed and leading to improved productivity.