Colorful Help is a scientific project located in the areas of computer science, Machine Learning, Deep Learning and assistive technology, which aims to help people with visual impairments to perform routine tasks through a device capable of recognizing objects and their respective colors. Considering that independence is an important factor for self-esteem, it is recognized that, even if these people can rely on the sense of touch or existing mobile apps, this process can often be uncomfortable or complicated. Therefore, the intention of this research is to contribute to this audience by developing a device that the person can wear as an accessory, initially recognizing garments and communicating to the user, through audio, the shape and color of the clothes. For this, several Convolutional Neural Networks (CNN's) will be made to compare their performance and find out which one is capable of distinguishing and classifying the patterns found in the shapes of each object. Color distinction also compares a computer vision algorithm programmed in Python language and an RGB (Red, Green, Blue) color sensor, programmed in Arduino, and the intention is to find out which of the two methods has the best result and under what circumstances it is recommended. The first evaluation of the neural network will be done with your Test Dataset, which is a set of images similar to those used for your training. Her second evaluation and the computer vision algorithm test were performed with samples captured with a webcam, in order to ascertain the accuracy in a real situation, based on the average of correct answers. The evaluation of the RGB sensor, however, because it cannot recognize images, was done in a different way, just to evaluate its accuracy in readings. Thus, based on the acquired data, it was found that the elaboration of the device is feasible, as the most capable model among neural networks can predict shapes and colors at a distance of 650mm from the object, with an average accuracy of 95% in the 4 categories , while the computer vision algorithm achieved 90% average accuracy in the 9 colors of its programmed spectrum. As future forecasts, it is intended to implement more categories, precision in more saturated environments and/or with a lot of brightness variation, in addition to allowing the distinction of more details, such as differentiating a checkered outfit from a striped one, for example. Keywords: Visual impairment, Assistive Technology, Deep Learning, Machine Learning, Artificial Intelligence