Robin D. Tillett

Using model-based image processing to track animal movements

Abstract Previous work has shown how a trainable flexible model (a point distribution model) can be used to locate pigs in images. This paper extends the idea to tracking animal movements through sequences of images where a single pig is viewed from above. As well as position and rotation, more subtle motion such as bending and head nodding can be modelled. This type of model based tracking could be used to characterise animal behaviour over time. The technique was used on seven sequences and worked well in most cases. However, it is possible to lose lock, as happened in one sequence, and the method currently reported cannot restart tracking. Further developments are required to investigate model fitting methods and high level control over the fitting and tracking process.

Estimating Dimensions of Free-Swimming Fish Using 3D Point Distribution Models

Monitoring the growth of farmed fish is an important task which is currently difficult to carry out. An underwater stereo image analysis technique offers the potential for estimating key dimensions of free-swimming fish, from which the fish mass can be estimated. This paper describes the development of a three-dimensional point distribution model to capture the typical shape and variability of salmon viewed from the side. The model was fitted to stereo images of test fish by minimizing an energy function, which was based on probability distributions. The minimization was an iterated two-step method in which edges were selected for magnitude, direction, and proximity to the model, and the model was then fitted to the edges. A search strategy for locating the edges in 3D was devised. The model is tested on two image sets. In the first set 19 of the 26 fish are located in spite of their variable appearance and the presence of neighboring fish. In the second set the measurements made on 11 images of fish are compared with manual measurements of the fish dimensions and show an average error in length estimation of 5%.

Real-Time Segmentation of Plants and Weeds

This paper describes a real-time implementation of an algorithm to segment plants, weeds, and soil in images used to control an autonomous crop protection vehicle. We show how the segmentation module fits in to the overall system and mention previous work on other modules. We then describe a segmentation algorithm and show how it is implemented on two digital signal processors (DSPs) controlled by a single transputer. We present timing information to show that the algorithm can provide data at 5 Hz, which is fast enough to control the vehicle.

Shape measurements of live pigs using 3-D image capture

A photogrammetric stereo imaging system was used to capture 3-D models of live pigs, and quantitative shape measurements were extracted from cross sections of the models. Stereo images were captured of 32 pigs, divided into high-lysine and low-lysine diet groups, and 3-D models were built from the images. Each pig was imaged once per week for 14 weeks. After slaughter, 10 of the pigs were dissected for muscle and fat measurements. A sequence of algorithms was applied to the 3-D models: differential geometry to reveal surface curvature features and detect the spine; manual landmark placement; fitting a curve to the spine; determining the vertical axis of the body; placing a slice plane across the abdomen close to the P2 position; extracting a cross section; and fitting a shape model to the cross section. Differential geometry revealed many qualitative features of the musculature. The spine was a line of minimum curvature along the back. The high-lysine pigs had higher height-to-width ratios and flatter backs than the low-lysine pigs. The dissected total muscle mass had a -0·66 correlation with the flatness-of-back shape parameter, and a 0·64 correlation with weight.

Fuzzy multicriteria decision-making for long cane pruning: A system for standard and complex vine configurations

Section I of this article presents the agronomic and technical purposes involved in vine pruning and the decision model used to simulate the pruners reasoning (fuzzy multicriteria decision making). Section II describes the method used for the choice of the best aggregation operator. Section III describes the model decision which is used to make automatic decision on complex vine configurations. Section IV presents the results: the model was performed on training set made of 140 standard pruning cases, 69.4% of similarity were observed between the decision model and the expert.

Applying mesh conformation on shape analysis with missing data

A mesh conformation approach that makes use of deformable generic meshes has been applied to establishing correspondences between 3D shapes with missing data. Given a group of shapes with correspondences, we can build up a statistical shape model by applying principal component analysis (PCA). The conformation at first globally maps the generic mesh to the 3D shape based on manually located corresponding landmarks, and then locally deforms the generic mesh to clone the 3D shape. The local deformation is constrained by minimizing the energy of an elastic model. An algorithm was also embedded in the conformation process to fill missing surface data of the shapes. Using synthetic data, we demonstrate that the conformation preserves the configuration of the generic mesh and hence it helps to establish good correspondences for shape analysis. Case studies of the principal component analysis of shapes were presented to illustrate the successes and advantages of our approach.

A stereo imaging system for the metric 3D recovery of porcine surface anatomy

We have succeeded in capturing porcine 3D surface anatomy in vivo by developing a high‐resolution stereo imaging system. The system achieved accurate 3D shape recovery by matching stereo pair images containing only natural surface textures at high (image) resolution. The 3D imaging system presented for pig shape capture is based on photogrammetry and comprises: stereo pair image acquisition, stereo camera calibration and stereo matching and surface and texture integration. Practical issues have been addressed, and in particular the integration of multiple range images into a single 3D surface. Robust image segmentation successfully isolated the pigs within the stereo images and was employed in conjunction with depth discontinuity detection to facilitate the integration process. The capture and processing chain is detailed here and the resulting 3D pig anatomy obtained using the system presented.

Image-guided robotics for the automation of micropropagation

Describes a robots workstation used as a test bed for evaluating tools and techniques for automating the technique of micropropagation. A number of novel plant manipulation tools have been developed. These tools are mounted on a Cartesian geometry robot driven by stepper motors to perform the required harvesting, cutting and planting movements. Overall robot direction is provided by a program which is updated according to the shape, position and orientation of a particular microplant. Information from a video camera can either be displayed on a monitor and assessed by an operator, or can be analysed by a vision processing system. This paper describes the robot workstation and discusses the various vision analysis tasks necessary for automatic robot guidance.<<ETX>>

Model-based image processing for characterizing pigs in scenes

This paper reports on the development of a model-based technique to locate pigs in fairly unstructured scenes. Model-based image processing is potentially a very powerful technique for identifying and classifying images. It is particularly relevant for biological objects such as animals which are difficult to define in numerical terms. Here the model is based on an image of a typical pig viewed from above . The model can then be rotated translated scaled and bent laterally to find a good match within an image of another pig. The output of the model helps to segment and classify the pig and the information can be used to guide further localized image processing. The development of the model and methods of fitting it to the image are described and results on a sample image are presented. 1.