Towards Fast Detection and Classification of Moving Objects

The detection and classification of moving objects are fundamental tasks in computer vision. However, current solutions typically employ two isolated processes for detecting and classifying moving objects. First, all objects within the scene are detected, then, a separate algorithm is employed to determine the subset of objects that are in motion. Furthermore, diverse solutions employ complex networks that require a lot of computational resources, unlike lightweight solutions that could lead to widespread use. We propose an enhancement along with an extended explanation of TRG-Net, a unified model that can be executed on computationally limited devices to detect and classify only moving objects. This proposal is based on the Faster R-CNN architecture, MobileNetV3 as a feature extractor, and an improved GMM-based method for a fast and flexible search of regions of interest. TRG-Net reduces the inference time by unifying moving object detection and image classification tasks, limiting the regions proposals to a configurable fixed number of potential moving objects. Experiments over heterogeneous surveillance videos and the Kitti dataset for 2D object detection show that our approach improves the inference time of Faster R-CNN (from 0.176 to 0.149 s) using fewer parameters (from 18.91 M to 18.30 M) while maintaining average precision (AP = 0.423). Therefore, the enhanced TRG-Net achieves more tangible trade-offs between precision and speed, and it could be applied to address real-world problems.