Predicting the Risk Level and Position of Objects in the Forklift's Blind Spot Area Using Artificial Neural Network

This paper presents a new application of artificial neural networks (ANN) for identifying blind spot condition severities and positions of a forklift. A microcontroller-based device is also developed for measuring the distance at the rear blind spot areas. The built ANN can also be deployed in the device. The training data is obtained from the measurement in a 5-ton capacity forklift, and the identification is determined based on the literature and visual checking from the operator's point of view. The input data include the distance and angle obtained from readings of ultrasonic sensors configured like a 180° radius radar. The predicted risk level predictions are categorized into safe, cautious, and dangerous. The detected object's position predictions are categorized into left, rear, and right. The neural network model employed is built based on a backpropagation algorithm. After varying the hidden node number, the best results were achieved at 70. For training and testing cases, both outputs' performance is in good agreement with expert determination identified by the precision, recall, and F1 score metrics, with these parameters yielding a value of 1. The results indicate that the developed system has promising potential for predicting blindspot severities.