Human detection through sensor fusion and convolutional neural networks : a thesis presented in fulfilment of the requirements for the degree of Master of Engineering in Mechatronics at Massey University, Manawatu, New Zealand
Loading...
Files
Date
2023
DOI
Open Access Location
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Massey University
Rights
The Author
Abstract
Safety practices in industrial environments can be improved by observing patterns of close calls or accidents. Machine detection of humans is one method that can be used to collect data for this purpose, but it has challenges achieving reliable results in complex and varied environments. Sonasafe, a company manufacturing and managing proximity safety systems for industrial environments, is looking to advance their capability to detect people. This thesis is concerned with object detection, with a specific focus on increasing the accuracy of human detection in industrial environments. Object detection requires the identification of specific, distinguishable features that can be distinctively associated with the object being detected, with the additional challenge of determining its location. The distinctiveness of an object’s features is determined by the type of object, its surroundings, and the sensor being used to detect the object. This study began with an evaluation of sensors that could contribute to distinguishing identifying human features and the mitigation of negative environmental conditions. The senses selected for testing were a standard visual camera for general feature recognition, an infrared camera for temperature features to mitigate the effects of dark environments, along with a millimetre wave (mmWave) radar to enhance location accuracy. This study reviewed the fusion of these sensors in Convolution Neural Networks to determine their respective performance in identifying different features when detecting a person in a complex environment. The study found that for a simple environment, with little noise and uniform data, single sensors had a higher rate of success, but that as the environment became more complex with more variability, the fusion of all three sensors produced more accurate results. Overall sensor fusion was seen to improve the accuracy of human detection in more complex industrial environments.