Implementation of Faster RCNN Algorithm for Smart Robotic ARM Based on Computer Vision

Robotic arm is a powerful mechanism used in industries to carry out pick and place tasks with great speed and accuracy. This task can either be accomplished by an arm that follows a preprogramed trajectory or an arm that can use its own intelligence to complete the task using Machine Learning. The problem with an arm which lacks intelligence is that with changing circumstances they cannot acclimatize well, the arm needs human intervention to reprogram as per requirement and perform only as per instructed. However, the intelligent robotic arm can be made capable enough to make its own decision for tasks. Some already prevailing projects in this domain are majorly based on sensors for the classification and detection of objects for pick and place tasks. The arm that uses a color-based sensor for detection majorly has a drawback that is less accurate in exact color detection of objects and detection time is also compromised and if sensors fail, the entire system will not perform the intended task. To address this problem, we propose using the Faster Region-Based Convolutional Neural Networks ML model to implement an arm capable of automatically sorting red and blue boxes by capturing their images with an OV5674 camera and processing them on a Raspberry Pi 3 B+ to accurately detect and classify red and blue boxes. The mechanical arm movement is ensured by a precise inverse kinematic model, which helps pick and segregate packets according to the color they possess. Both accuracy and real-time computing performance are good in the experimental findings of the Faster RCNN model. The precision we achieved by using the Faster RCNN model is 78.8% for detecting and sorting red and blue boxes. Hence, experimentation using Faster RCNN for smart robotic arm has successfully resulted in good outcomes.