Mouna Afif

Indoor objects and recognition present a very important task in artificial intelligence (AI) and computer vision fields. This task is an increasingly important especially for blind and visually impaired (BVI) indoor assistance navigation. An increasing interest is addressed for building new assistance technologies used to improve the daily life technologies used to improve the daily life activities qualities for BVI persons. To fulfill this need we propose in this work a new deep learning based techniques used for indoor wayfinding assistance navigation. we propose to use in this paper a new deep learning-based technique based on You Only Learn One Representation YOLOR network. This network enables a combination between implicit and explicit learning and knowledge just like the human brain can do. By introducing the implicit knowledge, the neural network is able to generate a unified representation that can serve for different tasks. In order to train and test the proposed indoor wayfinding assistance system, we proposed to work with the proposed indoor signage dataset. Based on the conducted experiments, the proposed indoor wayfinding system has demonstrated very interesting results. We applied different optimizations techniques in order to reduce the network size and parameters number to make the proposed model suitable for implementation on embedded devices. As a detection performance, we obtained 95.62% mAP for the original version of YOLOR network and 93.12% mAP for the compressed version and 28 FPS as detection speed.

Indoor objects detection and recognition present an active research axis in computer vision and artificial intelligence fields. Various deep learning-based techniques can be applied to solve the objects detection problems. With the apperance of deep convolutional neural networks (DCNN) a great breakthrough for various applications was achieved. Indoor objects detection present a primary task that can assist blind and visually impaired persons (BVI) during their navigation. However, building a reliable indoor objects detection system used for edge devices implementations still presents a serious challenge. To address this problem, we propose in this work to build an indoor objects detection system based on DCNN network. Cross-stage partial network (CSPNet) has been used for the detection process and a lightweight backbone based on EfficientNet v2was used as a network backbone. In order to ensure a lightweight implementation of the proposed work on FPGA devices, various optimizations techniques have been applied to compress the model size and reduce its computation complexity. The proposed indoor objects detection system was implemented on a Xilinx ZCU 102 board. Training and testing experiments have been conducted on the proposed indoor objects dataset that count 11000 images containing 25 landmark classes. The proposed work achieved very competitive results in terms of detection accuracy and processing time for the original CSP-EfficientNet v2 network as well as for the proposed compressed version.

Since 2019, COVID-19 disease caused significant damage and it has become a serious health issue in the worldwide. The number of infected and confirmed cases is increasing day by day. Different hospitals and countries around the world to this day are not equipped enough to treat these cases and stop this pandemic evolution. Lung and chest X-ray images (e.g., radiography images) and chest CT images are the most effective imaging techniques to analyze and diagnose the COVID-19 related problems. Deep learning-based techniques have recently shown good performance in computer vision and healthcare fields. We propose developing a new deep learning-based application for COVID-19 segmentation and analysis in this work. The proposed system is developed based on the context aggregation neural network. This network consists of three main modules: the context fuse model (CFM), attention mix module (AMM) and a residual convolutional module (RCM). The developed system can detect two main COVID-19-related regions: ground glass opacity and consolidation area in CT images. Generally, these lesions are often related to common pneumonia and COVID 19 cases. Training and testing experiments have been conducted using the COVID-x-CT dataset. Based on the obtained results, the developed system demonstrated better and more competitive results compared to state-of-the-art performances. The numerical findings demonstrate the effectiveness of the proposed work by outperforming other works in terms of accuracy by a factor of over 96.23%.

Traffic sign detection is one of the most important tasks for autonomouspublic transport vehicles. It provides a global view of the traffic signs on the road. Inthis chapter, we introduce a traffic sign detection method based on auto-encodersand Convolutional Neural Networks. For this purpose, we propose an end-to-endunsupervised/supervised learning method to solve a traffic sign detection task. Themain idea of the proposed approach aims to perform an interconnection between anauto-encoder and a Convolutional Neural Networks to act as a single network to detecttraffic signs under real-world conditions. The auto-encoder enhances the resolution ofthe input images and the convolutional neural network was used to detect and identifytraffic signs. Besides, to build a traffic signs detector with high performance, weproposed a new traffic sign dataset. It contains more classes than the existing ones,which contain 10000 images from 73 traffic sign classes captured on the Chinese roads.The proposed detector proved its efficiency when evaluated on the custom dataset byachieving a mean average precision of 86.42%.

Building new systems used for indoor assistance navigation and wayfindingin indoor places present a crucial and primary step to contributing to smart indoorenvironments. Indoor objects recognition and classification using deep neural networks(DNNs) present very powerful tools to assist blind and sighted persons during theirindoor navigation. This chapter proposes to develop a new indoor assistance navigationsystem using deep convolutional neural networks. The proposed system was evaluatedusing different types of deep learning-based models. The developed system can behighly recommended to contribute to a smart environment and to be applied for smarthomes applications. Experiments conducted in this work have shown the efficiency andthe robustness of the developed indoor object classification system. Experiment resultsobtained are very competitive in terms of classification rates which come up to 99.9%.