Radar-based vision algorithms for in cabin monitoring applications
In recent years, the number of infant deaths being left inside a vehicle has considerably increased, most of them due to careless parents unknowingly leaving the infant locked in the car. Therefore, this thesis aims to put forward a solution preventing this kind of infant deaths by detecting human p...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2023 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/386168 |
| Acceso en línea: | https://hdl.handle.net/2117/386168 |
| Access Level: | acceso abierto |
| Palabra clave: | Deep learning Artificial intelligence Deep Learning AI Child Presence Detection Agorithm Inteligencia Artificial Algoritmo Aprenentatge profund Intel·ligència artificial Àrees temàtiques de la UPC::Enginyeria de la telecomunicació::Processament del senyal Àrees temàtiques de la UPC::Informàtica::Intel·ligència artificial |
| Sumario: | In recent years, the number of infant deaths being left inside a vehicle has considerably increased, most of them due to careless parents unknowingly leaving the infant locked in the car. Therefore, this thesis aims to put forward a solution preventing this kind of infant deaths by detecting human presence inside the vehicle, especially infants and babies. In other words, a solution based on the installation of a Child Presence Detection (CPD) system inside the vehicle. This preventing system is based on different radar technologies together with the use of Deep Leaning techniques, which use multiple layers to extract relevant features of the data gathered by the radar and produce an alarm if necessary. This consists of an alternative low-cost non-contact implementation by using Doppler and FMCW radars, which can easily be integrated in the car structure while delivering high accuracy measurements and presenting lower power consumption. From the multiple Deep Learning structures that exist, this thesis explores the most interesting ones dealing with this kind of radar data, such as CNN, RNN, R-CNN among many others. The implementation using the 24GHz Doppler radar has been able to perform CPD with a considerably high accuracy considering it is a low-cost solution. On the other hand, the higher bandwidth of the 60GHz FMCW radar, not only has allowed to perform CPD, but it has also been able to localize the different target positions in the car. |
|---|