Identification and recognition of objects
Web and mobile applications using the best technology in objects reading.
About the project
The identification, recognition and reading of texts make up an area within Computer Vision that has been under development in the last decade, based on neural network models capable of detecting objects in moving images in real time, under adverse conditions in terms of lighting, rain, dust and other external elements…
The identification of these objects using a mobile device allows the recognition and verification process to be brought closer to interesting scenarios, with optimal battery consumption due to the low network traffic associated with it. The process is ad-hoc and no fixed control points are required. Some of its most relevant characteristics are the following:
- The process is carried out completely autonomously on the device, with a neural network model optimized to take advantage of the hardware accelerators incorporated in modern smartphones, allowing you to perform multiple detections on moving objects with a response time of less than 80 miliseconds.
- The system is capable of detecting small objects, in environments with low lighting, with different perspectives and high levels of occlusion. The process is completed in 5 stages: Detection, segmentation, optical character recognition, post-processing and verification.
- During detection, multiple video images are processed and detected objects in the scene are framed with a certainty indicator. From the selection the area of interest is cut out and the characters are segmented.
- It is possible to perform optical character recognition on the objects detected from a process of cropping and segmentation by areas of interest within the original image/ frame.
The effectiveness of recognition is around 90% of successful cases in worst case scenarios.
The stream is done in real time, as the system is capable of processing multiple video images in a range of 40 to 60 fps. The deformations that arise from the difference of the scene and camera planes are effectively corrected.
One of the most relevant aspects to highlight is that the entire process has been optimized to improve the performance of the device’s battery, drastically reducing its consumption.