WCSE

Real Time Object Detection for Visually Impaired Person Using Tensor Flow Lite

Abstract— Real Time Object Detection Using Tensor Flow Lite system has been developed to help visually impaired people for navigation and surrounding objects detection. This system is based on raspberry pi, a single board compute model and Tensor Flow lite framework. The algorithm developed is tested for detecting objects like a table, a chair, a TV, a laptop, a mouse, a cell phone, a bottle etc. This system is capable of detecting people as well as objects. The detection accuracy of 70% is achieved. The testing is done in varying light, background, and distance in indoors as well as outdoor scenarios. This system uses Google based sample quantized SSDLite-MobileNet-v2 object detection model, which is trained of the MSCOCO dataset and converted to run on TensorFlow Lite. The information regarding the detected object is converted into audio for guiding the visually challenged person

Index Terms—Tensor flow lite, MS-coco, Raspberry pi, object detection, gtts

Aditi Hemant khandewale, Vinaya V. Gohokar, Pooja Nawandar
MIT WORLD PEACE UNIVERSITY, INDIA

[Download]

Cite: Aditi Hemant khandewale, Vinaya V. Gohokar, Pooja Nawandar , " Real Time Object Detection for Visually Impaired Person Using Tensor Flow Lite " Proceedings of 2020 the 10th International Workshop on Computer Science and Engineering (WCSE 2020), pp. 150-156, Shanghai, China, 19-21 June, 2020.