LSTM-based Traffic Gesture Recognition using MediaPipe Pose

An important aspect of Autonomous Vehicles (AV) is having the ability to replicate and surpass a human's perception through the use of sensors. This will allow the AV to view its surroundings and make proper decisions during operation to guarantee the safety of both its passengers and surroundings. To achieve this, a pose extraction algorithm called MediaPipe, and an LSTM classification model will be used in conjunction to determine and classify traffic gestures from video input. It was seen that the proposed model was able to achieve good results based on a data set consisting of traffic gestures performed in an indoor environment, with a stationary camera. The proposed key point detection and LSTM model approach is a promising solution to the traffic gesture classification problem.