We study active object tracking, where a tracker takes visual observations (i.e., frame sequences) as input and produces the corresponding camera control signals as output (e.g., move forward, turn left, etc.). Conventional methods tackle tracking and camera control tasks separately, and the resulting system is difficult to tune jointly. These methods also require significant human efforts for image labeling and expensive trial-and-error system tuning in the real world. To address these issues, we propose, in this paper, an end-to-end solution via deep reinforcement learning. A ConvNet-LSTM function approximator is adopted for the direct frame-to-action prediction. We further propose an environment augmentation technique and a customized reward function, which are crucial for successful training. The tracker trained in simulators (ViZDoom and Unreal Engine) demonstrates good generalization behaviors in the case of unseen object moving paths, unseen object appearances, unseen backgrounds, and distracting objects. The system is robust and can restore tracking after occasional lost of the target being tracked. We also find that the tracking ability, obtained solely from simulators, can potentially transfer to real-world scenarios. We demonstrate successful examples of such transfer, via experiments over the VOT dataset and the deployment of a real-world robot using the proposed active tracker trained in simulation.

End-to-End Active Object Tracking and Its Real-World Deployment via Reinforcement Learning

End-to-end Active Object Tracking and Its Real-world Deployment via Reinforcement Learning

In this study, a real-time system design, which can track (RGB) targets in dynamic environments with an active camera, was implemented. Object tracking applications are quite important for military, surveillance system and operational robot applications and getting more important day by day. This design allows us tracking an object using fewer cameras. The design consists of three main parts that are object detection, mapping, tracking the object. When the camera catches the object, first it detects the shape of the object and creates bounding box, according to bounding box information, the algorithm calculates the centroid of the object. Object coordinates are determined using centroid of the object then tracking process works by activating motors via Arduino-MATLAB communication. The motors placed on a platform called pan-tilt platform. The platform can turn 270 and 180 degrees on×and y-axis respectively.

Color based moving object tracking with an active camera using motion information

We present an improvement in visual object tracking and navigation for mobile robot implementing the advantage actor- critic (A2C) reinforcement learning architecture on top of the Gym-Gazebo framework. This work provides an easier way to integrate reinforcement learning algorithms for navigation and object tracking tasks in robotics field. We train the convolutional-recurrent model employed for the policy estimation in an end-to-end manner. The robot is able to follow a simulated human walking in an indoor environment by using the sequence of images provided by the robot camera. The input of the algorithm is acquired and processed directly in ROS-Gazebo environment. The policy learned by the robot agent proved to generalize well also in an environment with different size and shape with respect to the training one. Moreover, the policy allows the robot to avoid obstacles while following the tracking target. Thanks to these improvements, we can straightforwardly apply the tracking system in a real world robot for a person following task in indoor environments.

Supporting Impaired People with a Following Robotic Assistant by means of End-to-End Visual Target Navigation and Reinforcement Learning Approaches

This survey reports the approaches for object detection and successful target tracking based on image acquisition for real time applications. Moving target detection and its tracking is an important research topic in computer vision and robotics. Complexity of algorithms for this purpose ranges from simple, single and stationary targets tracking to complex, multiple and moving targets handling. Their applications range from basic object detection and pick & place task using a simple robotic arm to complex surveillance using Unmanned Air Vehicles (UAVs). Depending upon the application scenario, various algorithms have been proposed. The objective of this research is to present a comprehensive survey on reported multipurpose algorithms for target real time detection and tracking. Particular consideration has been paid to the underlying norms of each algorithm and its ability to tackle various situations. The paper, besides introducing relevant algorithms and key terminologies, presents a comparative analysis of various associated techniques. It is anticipated that the present review will provide a very useful framework for researchers working in the area of computer vision. The review, in addition to helping them in choosing a particular algorithm, will offer guidance to tailor an existing technique for a specific problem in hand. (Javed Iqbal, Mustafa Pasha, Riaz-un-Nabi, Hamza Khan, Jamshed Iqbal. Real-Time Target Detection and Tracking: A Comparative In-depth Review of Strategies. Life Sci J 2013; 10(3): 804-813). http://www.lifescience site.com 121

/pdf/real-time-target-detection-and-tracking-a-comparative-in-7nt6vsz3p1.pdf

Real-Time Target Detection and Tracking: A Comparative In-depth Review of Strategies

This paper presents a vision-based navigation strategy for a pan and tilt platform and a mounted video camera as a visual sensor. For detection of objects, a suitable image processing algorithm is used. Moreover, estimation of the object position is performed using the Kalman filter as an estimator. The proposed method is implemented experimentally to a laboratory-size pan and tilt platform. Experimental results show good target tracking by the proposed method in real-time.

Bahare Torkaman

Papers

Real-time visual tracking of a moving object using pan and tilt platform: A Kalman filter approach