While human drivers intuitively respond to the gestures of traffic officers, autonomous vehicles do not inherently possess this capability, necessitating the development of sophisticated road traffic control gesture recognition systems. Recognizing gestures is crucial for the seamless integration of autonomous vehicles into human environments. Present-day approaches generally integrate various modalities, including bone vectors, keypoints, and picture characteristics. On the other hand, our neural network architecture uses only body skeleton data to generate state-of-the-art outcomes. We use an improved Bidirectional Long Short-Term Memory (BiLSTM) model that has been carefully optimized for best results. We extensively train and assess our model on the TCG dataset-which was created especially for real-world applications-using 3D body positions across time. This method demonstrates significant advancements, improving accuracy by over 5% compared to prior models, underscoring the robust potential of our gesture recognition system for autonomous vehicles.
Enhanced BiLSTM for Traffic Police Gesture Recognition in Autonomous Vehicles
2024-12-07
636012 byte
Conference paper
Electronic Resource
English
TRAFFIC POLICE GESTURE RECOGNITION BY POSE GRAPH CONVOLUTIONAL NETWORKS
British Library Conference Proceedings | 2020
|