Rapid and accurate counting and recognition of flying insects are of great importance, especially for pest control. Traditional manual identification and counting of flying insects is labor intensive and inefficient. In this study, a vision-based counting and classification system for flying insects is designed and implemented. The system is constructed as follows: firstly, a yellow sticky trap is installed in the surveillance area to trap flying insects and a camera is set up to collect real-time images. Then the detection and coarse counting method based on You Only Look Once (YOLO) object detection, the classification method and fine counting based on Support Vector Machines (SVM) using global features are designed. Finally, the insect counting and recognition system is implemented on Raspberry PI. Six species of flying insects including bee, fly, mosquito, moth, chafer and fruit fly are selected to assess the effectiveness of the system. Compared with the conventional methods, the test results show promising performance. The average counting accuracy is 92.50% and average classifying accuracy is 90.18% on Raspberry PI. The proposed system is easy-to-use and provides efficient and accurate recognition data, therefore, it can be used for intelligent agriculture applications.

Video anomaly detection aims to detect the segments containing abnormal events from video sequence, which is a current research hotspot due to the importance in maintaining social security. Recent detection methods tend to build frame reconstruction or frame prediction model based on deep learning to learn features of events. The reconstruction-based methods reproduce the input frame one-to-one, inevitably losing some temporal features. The prediction-based methods predict frames according to the natural time order, but ignore the reverse time information, causing the deviation in information learning. Besides, anomaly evaluation methods based on patch-level error neglect the diversity of object sizes in complex scenes, and it is difficult to determine the optimal size of the error patch accurately. For these issues, we propose a bidirectional spatio-temporal feature learning framework with multi-scale anomaly evaluation strategy. A video sequence is input to a double-encoder double-decoder network, and bidirectional spatio-temporal features are obtained for bidirectional prediction by fusing forward and backward features extracted from the two encoders. The multi-scale anomaly evaluation method is implemented based on error pyramid and mean pooling, which effectively detects target objects with different sizes. Experiments on several publicly video datasets show that our method outperforms most of existing methods.

The assumption of a static environment is typical in many visual simultaneous localization and mapping (VSLAM) systems. However dynamic objects in open scenes will mislead feature association and even fail to match, which reduces the accuracy of localization. For dynamic scenarios, a robust visual SLAM system that utilizes weighted features, namely, named WF-SLAM is proposed in this paper, which is based on ORB-SLAM2. First, WF-SLAM applies the tightly coupled semantic and geometric dynamic target detection algorithm to obtain the dynamic information in the scene. Then, WF-SLAM defines feature point weights and initializes them with the dynamic information. Finally, the pose optimization in ORB-SLAM2 is changed to weight-based joint optimization. WF-SLAM significantly decreases mismatch and improves the accuracy of localization. Experiments are performed on the benchmark RGB-D dataset TUM and real-world scenarios, and the results demonstrate that WF-SLAM realizes significant improvements in term of localization accuracy compared to ORB-SLAM2 in dynamic environments and a more robust performance compared with state-of-the-art dynamic SLAM methods.

This paper presents a novel CPG-based gait generation of the curved-leg hexapod robot that can enable smooth gait transitions between multi-mode gaits. First, the locomotion of the curved leg and instability during the gait transitions are analyzed. Then, a modified Hopf oscillator is applied in the CPG control, which can realize multiple gaits by adjusting a simple parameter. In addition, a smooth gait switching method is also proposed via smooth gait transition functions and gait planning. Tripod gait, quadruped gait, and wave gait are planned for the hexapod robot to achieve quick and stable gait transitions smoothly and continuously. MATLAB and ADAMS simulations and corresponding practical experiments are conducted. The results show that the proposed method can achieve smooth and continuous mutual gait transitions, which proves the effectiveness of the proposed CPG-based hexapod robot control.