Two forensic instruments have been developed and evaluated with respect to basic characteristics. One instrument is used for automatic fiber detection and the other for breath alcohol measurement.

The fiber detection instruments utilized digital image processing to quantify the content of a reference color in images captured from the investigation material, i.e. tapes with attached textile fibers mounted on transparencies. The investigation material was mounted on a sample table having a motorized video camera for image acquisition. By aid of the HSI-color system (Hue, Saturation, Intensity) and streamlined analysis sequence it was possible to reach a working speed and color selectivity comparable to a human investigator. The analysis sequence was divided into two phases. In the first phase a fast direct color match was performed of an evenly distributed grid in the image. It was not necessary to process every picture element because the fibers were usually much longer than the diagonal of a captured video image. Images that were found having enough reference color in phase one were also processed in a second phase. This phase performed a more thorough image filtration to reduce noise and increase hit-quality. The filter rejected all parts of an image having no or little content of the reference color. Areas with large clusters, e.g. fibers, were on the other hand enhanced.

The performance was best for colors in the red region and worst for green colors. Very pale or dark colors were usually not possible to detect without penalties in decreased selectivity. It was also important to have samples with a fairly clean background without interfering particles such as soil and dust. Heavily aggregated fibers could also prevent accurate detection.

The breath alcohol instrument was based on the concept of an electronic nose. The main parts were an array of 10 to 14 gas sensors integrated in a gas flow system and a computer for control and sensor response acquisition. The setup was used to quantify the ethanol content in breath from individuals who had consumed ethanol-containing beverages of various quantities. Data evaluation and estimation of ethanol concentration were performed with multivariate methods such as projection to latent structures (PLS) and feed forward artificial neural networks (ANN).Gas chromatography was used in parallel to measure the actual ethanol content in each sample, thereby making it possible to generate learning sets to be used in the subsequent data evaluation.

In the final configuration ethanol could be quantified down to 2*10¹ mol ppm ethanol in gas phase. Investigation of the sensor array revealed properties, which had negative influence on the performance. Besides having slow responses (close to a minute) and recovery (between 6 and 7minutes) the array was also sensitive to humidity and varying environmental conditions. The slow recovery further imposed memory effects when analyzing several samples directly after each other.

In a forensic perspective the instrument for automatic fiber detection demonstrated such usefulness that it has been introduced on the market and acquired by several forensic laboratories. There are though plenty room for further enhancements, especially concerning the image processing in the second phase. The electronic nose has demonstrated new approaches to breath-alcohol measurement regarding both the sensory architecture and data evaluation. To be practically useful as a forensic method additional development is still needed. Besides improving or reconfiguring the sensor array, carrying out significant clinical testing, optimizing sampling and improving the gas flow system it will also be necessary to develop a reliable and easy handled calibration scheme that can be used during field operation.