An Efficient 3D Data Annotation and Object Detection Pipeline for Production Line

2D (dimensional) object detection algorithms are important in computer vision and scene understanding. However, lacking depth perception in 2D images and the availability of low-cost 3D sensors attracted researchers' attention to 3D object detection. The 3D object detection approach is more accurate due to its 3D understanding and less susceptible to lighting variations. Nevertheless, it also has challenges like expensive computation costs, lack of training data, cumbersome data collection process etc. In this work, we tried to address these challenges to some extent by proposing an efficient 3D object detection pipeline, with an easy-to-use 3D data annotation methodology. We have also explored synthetic 3D data creation techniques to enrich the data. Also developed and deployed a computationally inexpensive 3D active region generator on an edge device to reduce the overall computational cost of our proposed pipeline. Our trained model achieves around 94% accuracy while processing data in real-time.