CNN-based Passenger Detector for Public Transport Vehicles

Autonomous vehicle technology is becoming relevant not only for researchers but for industries as well. Many studies focus on autonomous cars as primary importance. The growing amount of research enables to improve the economy, safety and the environment. In this paper we want to give attention to the public transport system, more precisely to the bus and train systems. Our study is about the computer vision system and its integration into these vehicles with the aim to recognize passengers waiting at the bus stops or train stations. In particular, the network proposed follows the keypoint estimation approach, which uses a feature extraction backbone in order to generate a low resolution heatmap and, after detecting the peaks on it, regresses to the bounding box sizes and positions. The experimental results achieved an F-Score
$(\beta=2)$
value of 86% using a ResNet50 backbone.