Deep learning has revolutionized object identification and categorization, offering unrivaled efficiency and accuracy. This work chiefly focuses on the deep learning-based object recognition and classification method based on convolutional neural networks (CNNs). CNNs, a staple of modern computer vision tasks, are particularly useful for image data due to their ability to automatically learn hierarchical features. Here we present an enhanced method for recognizing objects in real-time from video streams and computing their measurements. It used OpenCV libraries to propose an object measurement method for real-time video that employs clever edge recognition, dilation, and erosion techniques. To this end, the proposed model employs performing in detection of objects using the image dataset trained up to date, allowing pixel values in a region to use edge detection, thus improving the bounding box of the object. Moreover, it proposes a solution to the problem using a convolutional neural network (CNN) for object detection algorithm. First, in CNN (5), it can be observed, the prediction with the corresponding ground truth indicate one example property. Next, a modulation fac-fore on the prediction error with respect to the case study attribute teen is introduced to modify the prediction error. Finally, we pass the ground truth and updated prediction to our loss function.