Machine vision represents a particularly attractive solution for sensing and detecting potential collision-course targets due to the relatively low cost, size, weight, and power requirements of vision sensors (as opposed to radar and Traffic Alert and Collision Avoidance System). This paper describes the development and evaluation of a real-time, vision-based collision-detection system suitable for fixed-wing aerial robotics. Using two fixed-wing unmanned aerial vehicles (UAVs) to recreate various collision-course scenarios, we were able to capture highly realistic vision (from an onboard camera perspective) of the moments leading up to a collision. This type of image data is extremely scarce and was invaluable in evaluating the detection performance of two candidate target detection approaches. Based on the collected data, our detection approaches were able to detect targets at distances ranging from 400 to about 900 m. These distances (with some assumptions about closing speeds and aircraft trajectories) translate to an advance warning of between 8 and 10 s ahead of impact, which approaches the 12.5-s response time recommended for human pilots. We overcame the challenge of achieving real-time computational speeds by exploiting the parallel processing architectures of graphics processing units (GPUs) found on commercial-off-the-shelf graphics devices. Our chosen GPU device suitable for integration onto UAV platforms can be expected to handle real-time processing of 1,024 × 768 pixel image frames at a rate of approximately 30 Hz. Flight trials using manned Cessna aircraft in which all processing is performed onboard will be conducted in the near future, followed by further experiments with fully autonomous UAV platforms. © 2010 Wiley Periodicals, Inc. © 2011 Wiley Periodicals, Inc.

/pdf/airborne-vision-based-collision-detection-system-3ro7q074bf.pdf

Airborne vision-based collision-detection system

Efficient training for dendrite morphological neural networks

The integration of unmanned aircraft into civil airspace is a complex issue. One key question is whether unmanned aircraft can operate just as safely as their manned counterparts. The absence of a human pilot in unmanned aircraft automatically points to a deficiency that is the lack of an inherent see-and-avoid capability. To date, regulators have mandated that an “equivalent level of safety” be demonstrated before UAVs are permitted to routinely operate in civil airspace. This chapter proposes techniques, methods, and hardware integrations that describe a “sense-and-avoid” system designed to address the lack of a see-and-avoid capability in UAVs.

/pdf/see-and-avoid-using-onboard-computer-vision-458o6pt8m3.pdf

See and Avoid Using Onboard Computer Vision

The track-before-detect processing technique has been employed in numerous computer vision based algorithms addressing the dim target detection problem. This processing technique has been shown to be effective under certain conditions; but in particularly noisy or highly cluttered environments, detection performance may be improved by introducing an image preprocessing stage to enhance the raw sensor measurements prior to integration. In this paper, we compare the 'Close-Minus-Open' (CMO) and 'Preserved-Sign' (PS) morphological image preprocessing techniques for suppressing unwanted noise and emphasising target features in the measurement images. This investigation is motivated by the unmanned aerial vehicle "sense-and-avoid" application, where morphology-based filters have demonstrated a degree of success in the detection of small pointlike features that may correspond to collision-course aircraft. For completeness, we also briefly examine two well published track-before-detect temporal filtering techniques which may be combined with the morphological pre-processing to detect dim, sub-pixel sized targets. Results from our simulation studies show that the PS approach achieves a higher detection rate than the CMO approach.

/pdf/a-study-of-morphological-pre-processing-approaches-for-track-px171u68ho.pdf

A study of morphological pre-processing approaches for Track-Before-Detect dim target detection

A Study of Morphological Pre-Processing Approaches for Track-Before-Detect Dim Target Detection

It is widely accepted that the design of morphological filters, which are optimal in some sense, is a difficult task. In this paper a novel method for optimal learning of morphological filtering parameters (Genetic training algorithm for morphological filters, GTAMF) is presented. GTAMF adopts new crossover and mutation operators called the curved cylinder crossover and master-slave mutation to achieve optimal filtering parameters in a global searching. Experimental results show that this method is practical, easy to extend, and markedly improves the performances of morphological filters. The operation of a morphological filter can be divided into, two basic problems including morphological operation and structuring element (SE) selection: The rules for morphological operations are predefined so that the filter's properties depend merely on the selection of SE. By means of adaptive optimization training, structuring elements possess the shape and structural characteristics of image targets, and give specific information to SE. Morphological filters formed in this way become certainly intelligent and can provide good filtering results and robust adaptability to image targets with clutter background.

Automatic target detection by optimal morphological filters

A practical neural network model for morphological filtering and a simulated annealing optimal algorithm for the network parameters training are proposed in this paper. It is pointed out that the optimal designing process of the morphological filtering network in fact is the optimal learning process of adjusting network parameters (structuring element, or SE for short) to accommodate image environment. Then the network structure may possess the characteristics of image targets, and so give specific information to the SE. Morphological filters formed in this way become certainly intelligent and can provide good filtering results and robust adaptability to complex changing image. For application to motional image target detection, dynamic training algorithm is applied to the designing process using asymptotic shrinking error and appropriate network weights adjusting. Experimental results show that the algorithm has invariant property with respect to shift, scale and rotation of moving target in continuing detection of moving targets.

Nong Yu

Papers

Automatic target detection by optimal morphological filters

Morphological neural networks for automatic target detection by simulated annealing learning algorithm