Active Object Detection With Multistep Action Prediction Using Deep Q-Network

Abstract

In recent years, great success has been achieved in visual object detection, which is one of the fundamental tasks in the field of industrial intelligence. Most of existing methods have been proposed to deal with single well-captured still images, while in practical robotic applications, due to nuisances, such as tiny scale, partial view, or occlusion, one still image may not contain enough information for object detection. However, an intelligent robot has the capability to adjust its viewpoint to get better images for detection. Therefore, active object detection becomes a very important perception strategy for intelligent robots. In this paper, by formulating active object detection as a sequential action decision process, a deep reinforcement learning framework is established to resolve it. Furthermore, a novel deep Q-learning network (DQN) with a dueling architecture is proposed, the network has two separate output channels, one predicts action type and the other predicts action range. By combining the two output channels, the action space is explored more efficiently. Several methods are extensively validated and the results show that the proposed one obtains the best results and predicts action in real time.

Leveraging Active Perception for Improving Embedding-based Deep Face Recognition

“…For example, in [15] it is demonstrated that developing a deep learning system that also predicts the next best move for a robot, using reinforcement learning, can significantly improve the performance of object detection, where the viewing angle, occlusions and the scale of each object can have a significant effect on the object recognition accuracy. Similar observations were also reported by more recent works [16], [17], [18]. At the same time, it is worth noting that active perception approaches often allow for developing faster and more lightweight DL models, since models are trained in order to solve a simpler problem.…”

Offset Detection of Grate Trolley’s Side Plate Based on YOLOv4

“…In recent years, the neural network-based methods have been widely applied in intelligent transportation systems [13][14][15], intelligent video surveillance [16,17], automatic monitoring [18,19], and industrial inspection [20,21] fields. By improving the Faster R-CNN network model, Sun et al realized the detection of scratch, oil pollution, block, and grinning four kinds of wheel hub defects quickly and accurately [22].…”

Toward Intelligent Industrial Informatics: A Review of Current Developments and Future Directions of Artificial Intelligence in Industrial Applications

“…In [72], authors propose a novel fault diagnosis method for complex circuit board design using feed forward neural networks and support-vector machines that learns from repair history and localizes the root cause of a failure. In [73], X. Han et al formulate active object detection in industrial settings as a sequential action decision process, and apply a deep reinforcement learning framework, the deep Q-network (DQN) with dueling architecture to solve this formulation, by learning an optimal action policy. A deep neural network based two-stage automated approach for estimating the remaining useful life (RUL) of bearings in industrial machinery is proposed in [74], and in [75], authors model disassembly sequence planning as an NP-hard (non-deterministic polynomial-time hardness) many-objective problem, and solve this using the tensorial memetic algorithm that combines genetic computations with local search.…”