Abandoned Object Detection in Vehicle Cabin Environment

Abstract

Abandoned object detection in the vehicle cabin environment is pivotal for the convenience of passengers. Existing methods for abandoned object detection are either aimed at outdoor surveillance or rely too much on training data. This study aims to develop an efficient algorithm for detecting abandoned objects in a vehicle cabin environment that relies less on training data and generalizes to untrained objects. The main approach involves utilizing a simple background model to extract candidate abandoned objects, followed by feature extraction using MobileNetV2 pre-trained on ImageNet-k for accurate detection. By comparing features before and after human presence, false positive proposals are effectively filtered out. To make the proposed method suitable for embedded execution, optimizations are performed to improve efficiency. We evaluate the proposed method on common left-behind objects videos with a Jetson Nano device. The results prove the efficacy and efficiency of the proposed method. However, the study is still limited by the lack of research on dynamic cameras, low-contrast infrared images, and persistent shadows.