To deal with the challenges of object detection and object grabbing tasks for service robots, we propose an object detection and segmentation algorithm based on the deep convolutional neural network (CNN) and depth-first algorithm to achieve object grabbing by using an RGB-D camera and the six-degree-of-freedom robotic manipulator. Firstly, the improved particle swarm optimization (PSO) algorithm is proposed to calibrate and optimize the hand-eye system of the experimental platform, which is a mobile robot equipped with UR5 mechanical arm and Kinect V2 sensor. Then, we collect the environmental information by the camera, where the depth images restored by the joint bilateral filtering algorithm and the original color images are calibrated. Finally, a depth learning method is used to detect the objects, and the depth information is used to achieve objects segmentation. We complete object grabbing based on the estimated 3D coordinates, which has proven the practicability and effectiveness of our proposed grabbing framework.
With the gradual maturation of smart power grids and modern image processing technology, intelligent detection system based on substation is widely used. For the real-time requirements of substation intelligent inspection, this paper proposes the model of intelligent inspection of substation with light weight target detection. Firstly, the MobileNet network is introduced to build the lightweight backbone feature extraction network. Secondly, the module of multi-scale feature fusion is constructed to obtain more feature information. Finally, the scaling factor pruning model channel is introduced to build the model for lightweight target recognition. The proposed algorithm is compared with an existing algorithm in the simulation of the three datasets of Safety helmet (hardhat) wearing detect dataset, Hard Hat Workers dataset, Safety Helmet Detection dataset. Extensive experiments show that the proposed algorithm is more effective. The average recognition accuracy of the proposed algorithm is 0.932, and up to 38 frames per second are processed.
In this paper we present a novel method called color support Normal Iterative Closest Point (color NICP) to recursively align point clouds. First of all, the matching criterion of NICP is improved based on the uncertainty analysis on the feature points and L*a*b* color space. Secondly, a six-dimensional k-d tree based nearest neighbor search is used to match corresponding points between the clouds. Our algorithm takes advantage of not only the 3D structure, but also the texture information of color image. Extensive experiments show that our color NICP algorithm improves the robustness and precision of point cloud registration.
Human action recognition for videos has been applied extensively in man–machine interaction systems, video surveillance, virtual reality, and patient monitoring, which are still challenging problems in computer vision due to the complex backgrounds, changeable movement speeds, and different shooting scales with multiperspectives. To improve the robustness and accuracy of the recognition algorithm, many state-of-the-art methods have been proposed.
Human action recognition from videos is one of the research hotspots in the field of computer vision. As recognition algorithms have been innovated continuously, recognizing fewer categories of actions in a specific environment is not a significant challenge. However, for the videos captured in realistic scenes, where the problems such as camera movement, viewpoint change, and target occlusion are widespread, and the stability of a recognition system needs to be further improved.