In recent years, significant advances have been gained in visual detection, and an abundance of outstanding models have been proposed. However, state-of-the-art object detection networks have some inefficiencies in detecting small targets. They commonly fail to run on portable devices or embedded systems due to their high complexity. In this workpaper, a real-time object detection model, termed as Tiny Fast You Only Look Once (TF-YOLO), is developed to implement in an embedded system. Firstly, the k-means++ algorithm is applied to cluster the dataset, which contributes to more excellent priori boxes of the targets. Secondly, inspired by the multi-scale prediction idea in the Feature Pyramid Networks (FPN) algorithm, the framework in YOLOv3 is effectively improved and optimized, by three scales to detect the earlier extracted features. In this way, the modified network is sensitive for small targets. Experimental results demonstrate that the proposed TF-YOLO method is a smaller, faster and more efficient network model increasing the performance of end-to-end training and real-time object detection for a variety of devices.

Abstract-This

Object tracking is a challenging research task because of drastic appearance changes of the target and a lack of training samples. Most online learning trackers are hampered by complications, e.g., drifting problem under occlusion, being out of view, or fast motion. In this paper, a real-time object tracking algorithm termed “robust sum of template and pixel-wise learners” (rStaple) is proposed to address those problems. It combines multi-feature correlation filters with a color histogram. Firstly, we extract a combination of specific features from the searching area around the target and then merge feature channels to train a translation correlation filter online. Secondly, the target state is determined by a discriminating mechanism, wherein the model update procedure stops when the target is occluded or out of view, and re-activated when the target re-appears. In addition, by calculating the color histogram score in the searching area, a significant enhancement is adopted for the score map. The target position can be estimated by combining the enhanced color histogram score with the correlation filter response map. Finally, a scale filter is trained for multi-scale detection to obtain the final tracking result. Extensive experimental results on a large benchmark dataset demonstrates that the proposed rStaple is superior to several state-of-the-art algorithms in terms of accuracy and efficiency.