Through action recognition, we can know what behavior the person in the image.
This paper (Passenger Detection and Pose Recognition using Deep Neural Networks) proposes a action recognize method based on deep learning combined with human detector to implement a action recognition system for passengers in public transportation vehicles, and proposes an architecture for passenger counting.
-
We set up two cameras in the environment and use 2D and 3D CNN to recognize static poses and dynamic actions.
-
Posture Recognition (2D CNN) : Recognize postures that do not need to consider time information.
-
Action Recognition (3D CNN) : Recognize the continuous motion of passengers.
-
-
In order to understand the number of passengers in the environment, we use two cameras to count passengers that based on the detection results of detector and association method.
-
To achieve the goal, we built a neural network to solve the double counting problem caused by the same person appearing on two cameras.
python=3.7
pytorch=1.6.0
numpy=1.19
opencv=3.4.2
CUDA 10.0
cuDNN 7.4.1
Use this to create environments refer to here
conda env create -f environment.yml
- bb_match : Using camera1 bounding box information to predict camera2 bounding box place.
- pose_classification : Using a single-frame approach to classify the action, we have two categories: seated and standed.
- action_classification : Using a multi-frame approach to classify the action for temporal movements, we have four categories: sitting, standing up, seated and standed.
- action_detection : Combination multi-frame, single-frame and bounding box to classify the action and calculate the number of people.
pose_classification dataset : Using minbus,Bus look down,Bus side view.
bb_match dataset : Using minbus.
action_classification dataset : Using action_frames.
action_detection dataset : Using pose_classification Dataset and action_classification Dataset.
- People Detection and Pose Classification Inside a Moving Train Using Computer Vision
- Human activity monitoring for falling detection. A realistic framework
- Dual Viewpoint Passenger State Classification Using 3D CNNs
- DeepPose: Human Pose Estimation via Deep Neural Networks
- Human Pose Estimation Using Convolutional Neural Networks
For Passenger-Action-Recognition bugs and feature requests please visit GitHub Issues. For business inquiries or professional support requests please visit https://vision.ee.ccu.edu.tw/index.php.