Unlike CNN, or the deepest learning for that matter, which is known for its capability of generating higher conceptual features automatically, which in-turn gives a major boost to the classifier, in case of traditional machine learning applications, such features need to be hand crafted by SMEs.

As we may also understand from our experience working on CPU-based machine learning classifiers, their performance is affected by high dimensionality in data and the availability of too many features to apply to the model, especially with some of the very popular and sophisticated classifiers such as Support Vector Machines (SVM), which used to be considered state-of-the-art until some time ago. 

In this section, we will understand some of the innovative ideas drawing inspirations from different fields of science and mathematics that led to the resolution of some of the cited challenges above, to fructify the concept of real-time object detection in non-CNN systems.