Vision Systems & Image Processing: Home PageJASCO Research Ltd has extensive experience in image acquisition and analysis systems. Our company has developed numerous machine vision applications ranging from systems that monitor and classify passing naval vessels under widely variable lighting conditions to specialized scientific image enhancement and processing software. In support of the aquaculture research community and fish farming industry JASCO has developed vision-based applications for tracking and analyzing the swim trajectories of fish larvae to lead to better understanding of their feeding and has contributed to the development of a system whereby the biomass of fish in farming pens is assessed through automated stereoscopic video sizing of a significant sample of individual subjects.
Target Detection & Tracking
JASCO Research has been developing vision-based target detection and tracking software for many years, particularly in the field of three-dimensional mapping of the motion of densely distributed point-like targets from twin orthogonal views. Originally intended as a technique for automatically mapping fluid flow lines using particle tracers, this technology has successfully been adapted to the analysis of the motion of larval fish and crustaceans and other quasi-microscopic aquatic life.
These subjects are studied in a laboratory aquarium with a back-illuminated viewing arrangement that profiles the targets against a bright background, a technique known as Silhouette Video Photography, and are filmed by twin video recording systems through perpendicular sides. The image processing poses significant challenges in discriminating between the actual targets and spurious features such as suspended particles, as well as in locating consistently from frame to frame the centroid of minuscule subjects that may vary in size and opacity as they change orientation.
In more recent work the target detection algorithm has been extended to handle other types of organisms that are elongated rather than point-like and can flex and curl as they move, and for which an extremity of the body rather than the centroid must be tracked. Having established the projection co-ordinates of the targets in each of the views, the next problem is to correlate them on the basis of their common axis taking into account the ambiguity due to parallax and the possibility of overlaps, which may require a one-to-many matching to ensure that the true spatial location is not missed.
Lastly a predictive algorithm is used to track at once the multiple targets in three dimensions using a set of proprietary ranking rules to optimize the survival of correct trajectories. The resulting time histories can then be analyzed to study behaviourally relevant motion parameters such as prey searching and capture patterns, and how they are affected by ambient factors such as currents or turbulence.
Visual Classification & Measurement
JASCO Research has worked extensively under contract from defence agencies in the area of harbour and naval chokepoint surveillance by a variety of sensory technologies, machine vision being one of them. From this work the company has gained considerable experience in the automatic classification and identification of ships passing within the field of view of a camera, experience that can be applied equally well to the recognition of other transiting targets.
The capabilities of the algorithms and techniques developed in this context include operability over a wide range of ambient lighting situations, tolerance to rotation of the subject image due to slant of the travel path relative to the focal plane of the camera, ease of introducing new categories in the classification process, and ability to perform optical character recognition on hull markings even in conditions of moderate distortion, low contrast or partial obliteration of the symbols.
Our company has also participated to a significant extent in the development for the fish farming equipment industry of a non-contact measurement system for the assessment of biomass in an aquaculture pen from stereoscopic video images. In its original form the technique required that an operator identify four key sizing points on a given fish in each view of a vertical stereo pair without the benefit of any image quality enhancement; this laborious and error prone procedure had to be repeated for a sufficient number of fish in multiple frames to construct a significant statistical sample.
JASCO Research developed a second generation interface in which the operator only has to select two easily located points per fish in just one view, with the added convenience of working on adaptively enhanced images of much greater sharpness, and the rest of the sizing is done automatically. This major operational improvement required the application of machine vision techniques that included contrast enhancement, spatial frequency domain filtering, edge detection, shape correlation and parametric fitting of outlines based on pre-defined morphological templates for a given species of fish.
JASCO Research scientists are experienced in working with imaging systems operating outside the visible range in spectral regions that include the ultraviolet (UV-A to UV-C) and the near, mid-range and far (thermal) infrared. Over the course of various projects in which our company has taken part, these types of imagers have been used in applications ranging from gas detection to electrical discharge diagnostics. Even within the bounds of the visible spectrum, our creative thinking has led to the development of innovative solutions grounded in a thorough understanding of how colour space information is processed in digital imaging devices.
One such technology is a high-speed burst videography system based on the electronic separation of overlaid images recorded with strobed lights filtered in the primary colours, which is made viable by a specialized digital hue processing method that compensates for residual cross-talk. Another is a software procedure capable of isolating very precise wavelength bands in a standard digitized colour video stream independently of luminosity variations, which enables the detection of any number of chromatic features in a scene either in real time or at the post-processing stage.
The usefulness of this technique has been demonstrated in applications ranging from the aerial detection of colour variations in tree canopies due to parasite infestation, to the assessment of algal bloom areas beneath the sea surface, to the fly-past cataloguing of power line insulators having particular colour properties. These technologies are available for use in their present form or can be further developed and adapted to meet specific user requirements.