| Summary: | The advancement of Autonomous Mobile Robots (AMR) is vastly being discovered and applied to several industries. AMR contributes to the development of Artificial Intelligence (AI), which focuses on the growth of human-interaction systems. However, it is safe to say that mobile robots work closely in real-time and under changing surroundings; this creates limitations that may affect the efficiency of the application. Object detection comes in two different architectures: Single-stage detector and Two-stage detector. This research presents the experimental results of the two-stage detector, namely the Faster Region-based Convolutional Neural Network (Faster R-CNN). The experiment is applied to the SODA10M dataset, which consists of 20,000 labelled images. Extensive experiments are performed with parameters tuning the model’s configuration like labelling, iteration value, and model’s baseline for optimal results. The detection model is evaluated using the standard model evaluator of Mean Average Precision (mAP) to study the object detection’s accuracy. Overall findings achieve the highest mAP of 37.51%, which aligns with the original research of the dataset’s developer. Nevertheless, this project has identified the experiment’s limitations contributing to the accuracy value of imbalanced labelling, the training environment, and the dataset size. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.
|
|---|