Viola–Jones Method for Robot Vision Purpose: A Software Technical Review

Viola–Jones (VJ) algorithm holds a prominent and undeniable position in the computer vision field. Despite not being the most recent technology, it aids research to achieve advanced levels. Researchers can practically implement the built-in detector to establish various esteemed inventions. The well-known image processing software MATLAB and iOS also offer the VJ-based functions, ‘vision.CascadeObjectDetector’ and ‘CIDetector’ in their platforms. However, it may be difficult for researchers outside the machine-learning field to choose suitable detectors. Therefore, this study intends to present the process for creating the detectors and explore the fundamental concept of a VJ detector. The software technical performances of these methods in processing 12 images with 155 faces are reviewed. This study found that the self-trained VJ detector has the most reliable accuracy since it achieved a 0.994 detection rate and precision of 1. Yet, it used the longest execution time and only be the best option if the execution time was not a significant factor. Overall, the CIDetector has the best performance in computation efficiency. It has a 0.776 detection rate and 0.833 precision. Nevertheless, it might be unable to detect small faces or faces with occluded eyes. If the user does not want any faces to be undetected, the ‘vision.CascadeObjectDetector’ is a better option as it has a 0.999 detection rate and 0.815 precision.