Visually impaired people (VIPs) have trouble doing daily living activities. Individuals who are visually impaired (VIPs) make up an important percentage of the population. Available models detects one object at a time and are unable to assist VIP's in real time scenario. Improved deep learning attentional dense based object recognition framework for visually impaired people proposed and implemented using spider environment with python. The model implemented using ITM which creates more fused images, Feature maps created using attentional dense module. The network's spatial comprehension is enhanced by the usage of Region Proposal Networks (RPN) and Spatial Information Guided Convolution (SIG Conv). To get ROI data, perception-dependent ROI pooling is employed. The fully linked layer of the network then conducts object detection using the inputs provided. An improved Gannet Optimization algorithm (IGOA) is used to modify the framework's hyper parameters. The proposed model have a testing accuracy of 96.35%, F1 score of 96. 00 % and have the frame processing time of 0.083s which is the best among all other models.
Deep Learning Attentional Dense based Indoor Object Recognition for Visually Impaired People
22.11.2023
624993 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Traffic signal lamp recognition device for visually impaired people
Europäisches Patentamt | 2024
|