How do smart cameras recognise transparent and reflective objects? SingTown Technology's OpenMV solves competition challenges.

In robot competitions, vehicles or robotic arms often need to identify the position of target objects against complex backgrounds to accomplish path planning or pick-and-place tasks.

This project requires the identification of targets as translucent acrylic hemispherical covers, approximately 10 centimetres in diameter, with a 2-centimetre border at the base, placed centrally on the indoor floor (within a 4×4 metre area). There may be up to three targets, with fixed positions.

The background is a non-solid colour floor with simple patterns, which can easily cause reflections and lighting interference.

To achieve precise coordinate recognition and pose estimation, the team plans to utilise SingTown's OpenMV intelligent camera as the on-board vision module for real-time detection and positioning of the acrylic cover.

The system must output the central point information of the target in the image coordinate system and transmit the data to the small vehicle's main control system to achieve automatic navigation, path adjustment, or task triggering.

The robotics competition project utilises a transparent target detection system built upon SingTown Technology's OpenMV intelligent camera. Its computer vision and image recognition technology, through the integration of reflective feature detection, edge extraction, and morphological analysis algorithms, enables precise identification of semi-transparent acrylic covers under complex lighting conditions.

Due to the strong reflectivity and weak edge contours of acrylic material, the system first employs dynamic thresholding and background adaptive algorithms to suppress interference from ground patterns. It then utilises Sobel edge detection and morphological closing operations to extract the edge contours of transparent acrylic. To address reflections, the system leverages brightness distribution and highlight detection, effectively distinguishing genuine boundaries from specular reflections.

Upon completion of identification, the program automatically calculates the centroid coordinates (x, y) of the acrylic cover within the camera's field of view and outputs them via the serial port to the SingTown car's main control board.

When detecting multiple targets, the algorithm can filter based on target area or location, identifying the cover within the central area of the site as the primary target.

Actual testing indicates that this solution achieves a recognition success rate exceeding 93% in indoor environments with uniform lighting, with a positional error of less than 1.5 centimetres (with the camera positioned 30 centimetres above the ground).

SingTown's OpenMV embedded vision system operates at a high speed (frame rate above 20fps), making it suitable for scenarios requiring edge deployment such as robotics competitions, vehicle navigation, and robotic arm grasping.

Its advantages include no need for markers, strong adaptability to transparent objects, and the ability to stably output coordinate signals across different ground patterns, providing precise visual input for automatic control systems.