Abstract

This article introduces a straightforward method for detecting the positions of products on conveyors using computer vision. The method comprises two main steps: 1) Detecting the objects and their locations by matching their distinct templates on the images, and 2) Determining the orientations via a region of interest (ROI) extracted from the vicinity of the detected locations. The objective is to reduce processing time to enable real-time applications.

For demonstration, the detection of plier pieces is used throughout the paper. A hole on the body of the plier piece is selected as the feature for detection, and the area around it is chosen as the picking location. The Hough transform is performed to detect the center of the hole, then the orientation is determined by the picking area around the hole. The proposed method performs well in both cases where a single object and multiple objects appear in the images. The method required around 40-50ms to process an individual image (1600x1200 pixels) to detect both single and multiple objects using personal computers (PCs). To increase the flexibility of the system, the method was implemented on Raspberry Pi 4. The processing time was approximately 230-240 ms in such cases.

Additionally, a method to reduce the amount of calculation and improve the process's performance is proposed. In this method, only a small region along the object's moving path is used as the region of interest (ROI) during the detection process. The location of the region is fixed, so-called the anchor location, and the detection process is only performed inside this location. In this way, the processing time is reduced to around 3-4ms with PC and 10-12ms with Raspberry Pi 4. The results are promising for real-time applications.