Object detection and recognition: using deep learning to assist the visually impaired.

BACKGROUND: Deep learning systems have improved performance of devices through more accurate object detection in a significant number of areas, for medical aid in general, and also for navigational aids for the visually impaired. Systems addressing different needs are available, and many manage effectively the detection of static obstacles. PURPOSE: This research provides a review of deep learning systems used with navigational tools for the visually Impaired and a framework for guidance for future research. METHODS: We compare current deep learning systems used with navigational tools for the visually impaired and compile a taxonomy of indispensable features for systems. RESULTS: Challenges to detection. Our taxonomy of improved navigational systems shows that it is sufficiently robust to be generally applied. CONCLUSION: This critical analysis is, to the best of our knowledge, the first of its kind and will provide a much-needed overview of the field.Implication for RehabilitationDeep learning systems can provide lost cost solutions for the visually impaired.Of these, convolutional neural networks (CNN) and fully convolutional neural networks (FCN) show great promise in terms of the development of multifunctional technology for the visually impaired (i.e., being less specific task oriented).CNN have also potential for overcoming challenges caused by moving and occluded objects.This work has also highlighted a need for greater emphasis on feedback to the visually impaired which for many technologies is limited.

Augmented reality for narrow area navigation in jaw surgery: Modified tracking by detection volume subtraction algorithm.

BACKGROUND AND AIM: Jaw surgery based on augmented reality (AR) still has limitations in terms of navigating narrow areas. Surgeons need to avoid nerves, vessels, and teeth in their entirety, not just root canals. Inaccurate positioning of the surgical instrument may lead to positional or navigational errors and can result in cut blood vessels, nerve channels, or root canals. This research aims to decrease the positional error during surgery and improve navigational accuracy by reducing the positional error. METHODOLOGY: The proposed 2D/3D system tracks the surgical instrument, consisting of the shaft and the cutting element, each part being assigned a different feature description. In the case of the 3D position estimation, the input vector is composed of image descriptors of the instrument and the output value consists of 3D coordinates of the cutter. RESULTS: Sample results from a jawbone-maxillary and mandibular jaw-demonstrate that the positional error is reduced. The system, thus, led to an improvement in alignment of the video accuracy by 0.25 to 0.35 mm from 0.40 to 0.55 mm and a decrease in processing time of 11 to 14 frames per second (fps) against 8 to 12 fps of existing solutions. CONCLUSION: The proposed system is focused on overlaying only on the area to be operated on. Thus, this AR-based study contributes to accuracy in navigation of the deeper anatomical corridors through increased accuracy in positioning of surgical instruments.