Performance Analysis of Deep Learning Approaches on Traffic Controller Hand Gesture Detection for Autonomous Driving Cars

Traffic controller hand gesture detection and classification plays a vital role in traffic management due to the growing vehicles mobility in urban areas. Besides, there is a need to support in real-time autonomous cars operating in various traffic conditions and constraints. This paper presents the development of deep learning approaches for traffic controller hand gesture detection and classification using locally collected data and hyperparameter settings. Results show that Yolo V8 98% extremely performed, Resnet at 95%, SSD MobileNet 96% and Faster R-CNN attains 97% accuracy with an average difference of 2%. Further, results show that Yolo V8 attains 0.009 seconds detection speed as compared to other models developed. Overall, the results show promising results to integrate the models developed for autonomous cars intelligence capabilities. Research and practical recommendations are offered.