From 'Seeing' to 'Understanding': SingTown Technology Drives Visual Advancement

In the application of intelligent service robots, environmental perception and target recognition are key technologies for achieving autonomous interaction and task execution.

The robot needs to recognise multiple objects in complex indoor environments, such as room doors and their open or closed status, human gestures, QR codes, and typical objects in daily scenarios (e.g., tabletops, items, markers, etc.). To achieve this, service robot manufacturers plan to adopt SingTown Technology's OpenMV smart camera to build a multi-object recognition system based on machine vision. This system, installed on the robot's head or body, can capture images in specified directions in real-time according to master control instructions, identify target categories, dimensions, and colour features, and return the recognition results to the host for path decision-making, task triggering, or voice interaction responses. Additionally, it can be extended to recognise QR codes, enabling navigation positioning, task identification, or authentication, thereby helping robots achieve higher-level intelligent behaviours in areas such as daily life, security, and guided tours.

A multi-target recognition system developed by service robot manufacturers based on SingTown Technology's OpenMV intelligent camera enables service robots to achieve intelligent visual perception in complex scenarios. It integrates computer vision, AI image recognition, template matching, colour tracking, morphological analysis, and QR code decoding algorithms.

The system captures real-time images of the robot's forward field of view via the camera. Edge-side AI utilises an object detection model to identify object edges and shape features, enabling the determination of door positions and their open or closed status. Through colour segmentation and brightness variation analysis, it can recognise the colours, dimensions, and relative distances of various objects.

This robot can also recognise gestures. Under close-range conditions, the system, based on contour tracking and key point matching algorithms, can identify common actions (such as raising a hand, pointing, waving, etc.). Simultaneously, the built-in QR code recognition function enables the robot to identify navigation points or receive task instructions.

This system can be deployed at the edge, with its algorithm operating efficiently in embedded environments, achieving a frame rate of over 20fps and a recognition latency below 100ms.

The system can communicate with the master microcontroller via serial port, I²C, or CAN bus, outputting recognition categories, dimensions, colours, and status information to achieve intelligent linkage with the host.

The test results indicate that the solution developed based on the SingTown OpenMV intelligent camera achieves an identification accuracy of over 92% under varying indoor lighting conditions. It supports multiple task scenarios, such as home companion robots, security patrol robots, and guide robots, significantly enhancing the robot's environmental understanding and interaction capabilities.