Traffic monitoring solution to the third track of the AI-City Challenge. The goal of this challenge is to track vehicles across multiple cameras placed in multiple intersections spread out over a city. The project first focuses in solving multi-tracking in a single camera, using faster r-cnn for object detection and the kalman filter for tracking. Then the solutions is extended to multiple cameras using siamese networks and metric learning.
Study of the behaviour of different configurations of RetinaNet, Faster R-CNN and Mask R-CNN, using the Detectron2 framework. The evaluation is done both qualitatively and quantitatively on KITTI-MOTS, MOTSChallenge, Cityscapes and out of context datasets.
Study of the behaviour of different configurations of RetinaNet, Faster R-CNN and Mask R-CNN, using the Detectron2 framework. The evaluation is done both qualitatively and quantitatively on KITTI-MOTS, MOTSChallenge, Cityscapes and out of context datasets.
Image classifier using both classic computer vision techniques, such as Bag of Visual Words classifier using SVM and logistic regression, and deep learning techniques, such as MLPs and InceptionV3. Finally we designed our own CNN: TinyNet.
Query by example CBIR system for finding paintings matches in a museum database using color, texture, text and feature descriptors. The datasets used present different distortions in the images: background, noise, overlapping text boxes, color corruption and rotation.