Object detection in car cabin environment

• Main Author: Lee, Wai Yeong
• Other Authors: Yap Kim Hui
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/167499

In recent years, the field of artificial intelligence (AI) has seen substantial investment and promise. One application that garnered significant interest is the use of AI in autonomous vehicles (AVs), where advanced car cabin monitoring systems could enhance safety features and expedite development. These systems can detect various objects in the car, such as passengers, pets and objects, and track their movements in real-time to provide security alerts. Traditionally, video action recognition techniques were used for this task, but a newer approach using object detection has shown promise. However, many state-of-the-art (SOTA) models struggle to balance accuracy and efficiency. Thankfully, the EfficientDet model family has shown promising results. The family of models adopts a compound scaling method, which allows the model parameters to be scaled based on available computing resources. This project aims to leverage the EfficientDet model family to improve car cabin monitoring systems. To achieve this goal, careful dataset selection and deep learning optimization techniques were implemented. The results were satisfactory, with the EfficientDet-D3 achieving a COCO mAP score of 64.6 on the manually annotated Drive & Act dataset with a latency of 35.4ms.