Toddler monitoring system in vehicle by using artificial intelligence

Kok, Jia Quan (2021) Toddler monitoring system in vehicle by using artificial intelligence. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

	Text (24 Pages) Toddler monitoring system in vehicle by using artificial intelligence.pdf - Submitted Version Download (633kB)
	Text (Full text) Toddler monitoring system in vehicle by using artificial intelligence.pdf - Submitted Version Restricted to Repository staff only Download (3MB)

Abstract

Nowadays, the vehicle becomes the main transportation and the safety of toddler need to be concerned. There is much news regarding the death of a toddler in the vehicle, such as heatstroke and accident caused by parents' carelessness. This project is to develop a monitoring system which able to detect a toddler's presence in a car and classify the seatbelt condition of the toddler. The toddler monitoring system is responsible for monitoring toddler conditions to ensure the toddler is safe. The system will protect the toddler by detecting the seatbelt condition of the toddler and alert the driver when the toddler is left inside the car after the vehicle is off. The monitoring system is vision-based and using machine learning algorithm to detect toddlers and seatbelts. The environment inside the car is emphasized in the project. Nvidia Jetson Nano is chosen as the microcontroller due to its powerful performance. The Faster R-CNN, SSD-Mobilenet and SSD-Inception algorithm with Tensorflow is used and compared for the detection and classification. The Faster R-CNN is not applicable on Jetson Nano due to the high computational requirement and Jetson Nano unable to use it to process the image and hence the performance only compared from SSD-Mobilenet and SSD-Inception. From the result, it is observed that SSD-Inception gives better performance with 77.70% of accuracy, 97.92% of precision and 77.47% of recall when detecting the toddler while SSD-Mobilenet perform better when detecting the class of seatbelt with 82.98% of accuracy, 99.07% of precision and 77.37% of recall. From the observation, the performance of both neural networks are just slightly difference and hence the decisive factor is the frame per second (FPS) of neural network when running in real-time. The SSD-Mobilenet detecting in real-time with 8.5 FPS which faster than SSD-Inception with 5.7 FPS.

Actions (login required)

View Item