Search

With the continuous development and progress of society, age estimation based on deep learning has gradually become a key link in human-computer interaction. Widely combined with other fields of application, this paper performs a gradient division of human fall behavior according to the age estimation of the human body, a complete priority detection of the key population, and a phased single aggregation backbone network VoVNetv4 was proposed for feature extraction. At the same time, the regional single aggregation module ROSA module was constructed to encapsulate the feature module regionally. The adaptive stage module was used for feature smoothing. Consistent predictions for each task were made using the CORAL framework as a classifier and tasks were divided in binary. At the same time, a gradient two-node fall detection framework combined with age estimation was designed. The detection was divided into a primary node and a secondary node. In the first-level node, the age estimation algorithm based on VoVNetv4 was used to classify the population of different age groups. A face tracking algorithm was constructed by combining the key point matrices of humans, and the body processed by OpenPose with the central coordinates of the human face. In the secondary node, human age gradient information was used to detect human falls based on the AT-MLP model. The experimental results show that compared with Resnet-34, the MAE value of the proposed method decreased by 0.41. Compared with curriculum learning and the CORAL-CNN method, MAE value decreased by 0.17 relative to the RMSE value. Compared with other methods, the method in this paper was significantly lower, with a biggest drop of 0.51.

In recent years, the research of autonomous driving and mobile robot technology is a hot research direction. The ability of simultaneous positioning and mapping is an important prerequisite for unmanned systems. Lidar is widely used as the main sensor in SLAM (Simultaneous Localization and Mapping) technology because of its high precision and all-weather operation. The combination of Lidar and IMU (Inertial Measurement Unit) is an effective method to improve overall accuracy. In this paper, multi-line Lidar is used as the main data acquisition sensor, and the data provided by IMU is integrated to study robot positioning and environment modeling. On the one hand, this paper proposes an optimization method of tight coupling of lidar and IMU using factor mapping to optimize the mapping effect. Use the sliding window to limit the number of frames optimized in the factor graph. The edge method is used to ensure that the optimization accuracy is not reduced. The results show that the point plane matching mapping method based on factor graph optimization has a better mapping effect and smaller error. After using sliding window optimization, the speed is improved, which is an important basis for the realization of unmanned systems. On the other hand, on the basis of improving the method of optimizing the mapping using factor mapping, the scanning context loopback detection method is integrated to improve the mapping accuracy. Experiments show that the mapping accuracy is improved and the matching speed between two frames is reduced under loopback mapping. However, it does not affect real-time positioning and mapping, and can meet the requirements of real-time positioning and mapping in practical applications.

Aiming at the problem that the model of YOLOv4 algorithm has too many parameters and the detection effect of small targets is poor, this paper proposes an improved helmet fitting detection model based on YOLOv4 algorithm. Firstly, this model improves the detection accuracy of small targets by adding multi-scale prediction and improving the structure of PANet network. Then, the improved depth-separable convolution was used to replace the standard 3 × 3 convolution, which greatly reduced the model parameters without reducing the detection ability of the model. Finally, the k_means clustering algorithm is used to optimize the prior box. The model was tested on the self-made helmet dataset helmet_dataset. Experimental results show that compared with the safety helmet detection model based on Faster RCNN algorithm, the improved YOLOv4 algorithm has faster detection speed, higher detection accuracy and smaller number of model parameters. Compared with the original YOLOv4 model, the mAP of the improved YOLOv4 algorithm is increased by 0.49%, reaching 93.05%. The number of model parameters was reduced by about 58%, to about 105 MB. The model reasoning speed is 35 FPS. The improved YOLOv4 algorithm can meet the requirements of helmet wearing detection in multiple scenarios.