The research presents a Traffic Police Hand Gesture Recognition System for Indian autonomous vehicles which adopts the MoveNet Thunder model from TensorFlow. The main goal focuses on enabling autonomous vehicles to correctly recognize and respond to traffic police hand signals to enhance their movement through areas controlled by human operators. The system effectively detects three primary motions such as “Stop” paired with the “Turn Left” hand movement and the “Move Forward” direction in real time. The creation of a customized dataset included 8,000 photographs which were obtained from different environmental settings through innovative image enhancement strategies. Through dense and dropout layers the neural network reaches an accurate level of 89 % while preventing overfitting. The Haar cascade-based face detection system excludes gesture recognition from happening unless the officer's camera views their face directly thus reducing misidentifications. Testing for the model validation took place through the Carla simulator under different light, weather and traffic condition environments. The accuracy rate of the system stands high for typical gestures, but minor detection errors occur between gestures with similar visual characteristics indicating future system optimization needs. The study confirms it is possible to implement gesture recognition technology into automated driving systems which presents a promising method for Indian traffic control management systems.
Enhancing Autonomous Vehicle Navigation: Traffic Police Hand Gesture Recognition for Self-Driving Cars in India Using Movenet Thunder
2025-04-17
462974 byte
Conference paper
Electronic Resource
English
GESTURE RECOGNITION SYSTEM FOR AUTONOMOUS VEHICLE TRAFFIC CONTROL
European Patent Office | 2023
|