According to the requirement of recognizing traffic police gestures for driver assistance systems and intelligent vehicles, a universal model for dynamic traffic police gesture recognition is firstly introduced, of which can accurately present the spatial context (such as the relative lengths of skeletons, the angles between each skeleton w. r. t. gravity, and part features) of the traffic police gestures. Secondly, an architecture which can respectively extract spatial context and temporal features of dynamic traffic police gesture is proposed. Meanwhile, deep neural network and LSTM are introduced to build a high-resolution traffic police gestures recognizer (namely HRTPGR). At last, the open Police Gesture Dataset is used to train and test TPGR, and the experimental results show that the TPGR achieves a state-of-the-art accuracy with 98.7% for dynamic traffic police gestures recognition, and has strong anti-interference ability to light, background and gesture shape changes.
A High-Resolution Approach for Dynamic Traffic Police Gestures Recognition Based on Spatial Context and Temporal Features Fusion
27.07.2023
1520506 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Traffic Police Dynamic Gesture Recognition Based on Spatiotemporal Attention ST-GCN
Springer Verlag | 2023
|