Cipherstone Technologies can assist in the acqusition of photographic hardware for prototype or final product development. The company has experience from developing hardware requirement specifictions and matching these against the offerings of international hardware manufacturers such as CMOS image sensor manufacturers. Photographic hardware may include CMOS or CCD cameras, lenses, filters, lighting and reflective panels.
|Prior to any algorithm development, we need to acquire representative images or image sequences to allow optimisation of algorithm parameters and training of statistical classifiers that may be used in the computer vision software. Ideally, the image dataset should reflect all expected variation in object appearance or acquisition conditions. Typically, image data is collected and split into at least two disjunct sets, one set for training and one for testing. Furthermore, an experimental protocol is defined allowing unbiased testing.|
In a typical application, data collection is followed by acquisition of markup or groundtruth data. Images are manually or semi-automatically processed to collect reference points and other data such as object characteristics and object states. The reference points can then be used to measure the performance of e.g. automatic object or feature detectors on a large dataset. Cipherstone Technologies has developed software tools for cost-effective markup acquisition on large image datasets.