This paper introduces an advanced soft plastic waste sorting system that efficiently detects and picks soft plastic bags filled with soft plastic waste on a moving conveyor belt. The system consists of a depth camera, an ABB IRB120 robot arm, a finger gripper, a computer and a conveyor belt. The RAPID programming language is used to store and execute all the pick-and-place trajectories and other functions of this system are programmed in Python. The system employs a You Only Look Once (YOLO) V7 based object detection module, a Simple Online Realtime Tracking (SORT) algorithm for object tracking, and an improved height calculation module for safe target height checking. The system also achieves collision-free pick-and-place trajectory generation for the robot arm. This is crucial in the real-world waste sorting industry environment, where conveyor belts are typically laden with waste stacked upon and close to each other. Using a collision-free trajectory can ensure that the robot arm avoids collision with other surrounding waste on the conveyor belt during the pick-and-place operation of the target recyclable waste. This paper presents a comparative study of YOLOv7, YOLOv3, and Fast-RCNN object detection models. Experimental results show that the YOLOv7 has the highest mean average precision (mAP) of 99.3%. Furthermore, the lab experiment shows the efficiency of the system with the 96% successful sorting rate and 2.9-second cycle time for sorting    (approximately 20.7 bags per minute). This work offers an innovative, safe and efficient solution in the soft plastic waste sorting industry, contributing to sustainability and automation.