Multi-stage Contour Based Detection of Deformable Objects

Abstract

We present an efficient multi stage approach to detection of deformable objects in real, cluttered images given a single or few hand drawn examples as models. The method handles deformations of the object by first breaking the given model into segments at high curvature points. We allow bending at these points as it has been studied that deformation typically happens at high curvature points. The broken segments are then scaled, rotated, deformed and searched independently in the gradient image. Point maps are generated for each segment that represent the locations of the matches for that segment. We then group k points from the point maps of k adjacent segments using a cost function that takes into account local scale variations as well as inter-segment orientations. These matched groups yield plausible locations for the objects. In the fine matching stage, the entire object contour in the localized regions is built from the k-segment groups and given a comprehensive score in a method that uses dynamic programming. An evaluation of our algorithm on a standard dataset yielded results that are better than published work on the same dataset. At the same time, we also evaluate our algorithm on additional images with considerable object deformations to verify the robustness of our method.

Salient Features and Snapshots in Time: An Interdisciplinary Perspective on Object Representation

“…Similarly, related to the AC finding on relative importance of areas of high curvature, Ravishankar [26] found that it is easier for an artificial agent to recognise deformed objects by placing emphasis on the bending around points of high curvature. It is further compelling that a piece-wise continuous mathematical function is naturally segmented at its points of discontinuity; corners are discontinuities of the derivative of a function of one dimension; edges are discontinuities of the derivative of a function of two dimensions; while points of high curvature (maxima, minima and inflexion points) are points where the first derivatives are zero.…”

Object detection based on scale-invariant partial shape matching

“…To overcome local deformation, Belongie et al [21] introduced the shape context, which is an edge histogram according to lengths and directions for each point. Ravishankar et al [22] formed the k-segment group to approximate the outer contour of objects to handle local deformation. There are also salient contours shape descriptors, such as [23] and [12].…”

2D Mass-spring-like Model for Prediction of a Sponge’s Behaviour upon Robotic Interaction

“…It is hard to catalogue the literature about deformable objects, because it covers a wide range of aspects which can be combined to obtain good simulations for different situations (we need to start as general as possible, but without loosing the other ones from sight), starting with computer vision (which covers: identification, representation [17,15], classification, tracking [9]), simulation and manipulation (with applications mainly in robotics, medicine [12], computer graphics [16,7] and industry [19]). Sometimes the technique involves manually characterising the behaviour of a family of materials [18], sometimes the main focus is in topological information [20], sometimes they overlap across fields or get combined for new applications, like the work by Luo and Nelson [10] where visual tracking can provide haptic information, after a calibration phase of a FEM model links vision and haptic feedback.…”