ChineseTrafficPolicePose is a network that classify 8 kinds of Chinese traffic police commanding poses by analyzing visual information.

ChineseTrafficPolicePose 是一个仅依靠视觉信息区分８种**交警指挥手势的网络

This code runs under tensorflow 1.4, it's hard to build now because Tensorflow has changed it's API a lot. A pytorch version of police gesture recognizer is being maintained with pretrained models available at:

基于pytorch的、有预训练模型的版本：

https://doi.org/10.1016/j.neucom.2019.07.103

We publish the Police Gesture Dataset, which contains the videos of Chinese traffic police commanding gestures, and ground truth gesture labels for each video frame.

Police Gesture Dataset Download link: Google Drive

Notice: This gif is outdated. current version support prediction for FULL BODY, include legs. Check the videos in our dataset for examples of supported videos.

Watch Videos:

• Frame by frame detection - Youtube Video
• Realtime detection - Youtube Video

Environment

• Only support Python3
• Use Tensorflow with GPU support

Training

• Download keypoint dataset from AI Challenger (~20GB).
• Rename the downloaded 4 folders to "train", "test_a", "test_b", "val".
• Extract downloaded dataset to parameters.TRAIN_FOLDER. You may change the content of this parameter according to your path.
• Run python3 PAF_train.py to train the keypoint network.
• Download our Traffic Police Gesture dataset (~2GB) according to Dataset section.
• Extract .csv files to dataset/csv_train and dataset/csv_test.
• Extract .mp4 files to dataset/policepose_video.
• Run python3 PAF_detect.py dataset/policepose_video -a to parse videos to skeletal data.
• Run python3 rnn_train.py to train LSTM using labels from dataset/csv_train and skeletal data from ./dataset/gen/rnn_saved_joints.
• Run python3 rnn_detect.py -p to predict test videos using name list from dataset/csv_test and skeletal data from ./dataset/gen/rnn_saved_joints.
• Run Python3 rnn_detect.py -e to print Edit Distance of predicted labels with ground truth labels from dataset/csv_test.