Computer Vision software for image and video identification

Computer vision often detects and locates objects in digital images and videos. As living organisms process images with their visual cortex, many researchers have taken the architecture of the mammalian visual cortex as a model for neural networks structured to perform image recognition.

Over the past 20 years, progress in computer vision has been remarkable. Some computer vision systems achieve 99% accuracy, and some run decently on mobile devices. Today's best image classification models can detect diverse catalogues of objects at high definition resolution in colour. Additionally, people sometimes use hybrid vision models that combine deep learning with classical machine-learning algorithms and perform specific sub-tasks.

Apart from image classification, other vision problems have been solved with deep learning, including image classification with localisation, object detection and segmentation, image reconstruction, colourisation and synthesis.

How does computer vision work?

Computer vision algorithms generally rely on neural networks or CNNs. CNNs use convolutional, pooling, ReLU, and loss layers to boost a visual cortex. In a fully connected layer, neurons are connected to all activation from the previous layer. Using a Softmax or cross-entropy loss for classification, it computes how the network training penalises the deviation between the predicted and actual labels.

Computer vision Applications

Google's retired self-driving car project, Waymo runs tests on seven million miles of public roads and the ability to navigate safely in daily traffic. However, it has faced at least one accident.

Tesla has three self-driving car models until today. After a mishap in 2018 by a Tesla SUV, the software has been upgraded.

Amazon Go stores have checkout-free self-service retail stores. Their in-store computer vision system identifies when shoppers pick up or return stock products. Shoppers are detected by and charged through an Android or iPhone app. If the Amazon Go software misses an item to count, the shopper can keep it for free. On the other hand, if the software falsely registers an item taken, the shopper avails the facility of flagging the item and gets a refund for that charge.

In healthcare, there are quite a few vision applications to classify certain features in pathology slides, chest scans, and others. These have demonstrated value when compared to proficient human practitioners and regulated approval.

There are also productive vision applications for agriculture such as agricultural robots, crop and soil monitoring, banking like fraud detection, document authentication, and remote deposits, and industrial monitoring such as private wells, site security, and work activity.

Computer vision frameworks and models

There are also productive vision applications for agriculture such as agricultural robots, crop and soil monitoring, banking like fraud detection, document authentication, and remote deposits, and industrial monitoring such as private wells, site security, and work activity.

Microsoft Computer Vision API can detect objects from a catalogue of 10,000, with 25 languages of labels. It also gets back bounding boxes for identified items. The face API can run in the cloud or on edge in containers. IBM Watson Visual Recognition has the capability of classifying images from a pre-trained model. It also allows you to train custom image models with transfer learning, perform object detection with object counting and train for visual inspection. Watson Visual Recognition can run through in the cloud, or on iOS devices applying Core machine learning.

As data analysis package Matlab has an optional Computer Vision Toolbox and capacity of integration with OpenCV, it can perform image recognition using deep learning and machine learning.

As we have seen, computer vision systems have become efficient enough to be useful and more accurate than humans in some cases. Applying transfer learning, customisation of vision models has turned practical for mere mortals. Computer vision is no longer the exclusive domain of PhD-level researchers.