Search

In order to improve the recognition rate of multi-face image, this paper proposes a face image recognition method based on haar-like and Euclidean distance. First of all, the initial features of Haar are moved in the image and gradually enlarged, the pixel sum of the feature area is obtained quickly by using the integral graph. The Haar feature is obtained by calculating the difference between black and white pixels in the feature region, and the threshold of weak classifier is calculated. Then, the extracted facial feature data is trained and classified to form a cascade classifier for face detection. Finally, the face of the image to be detected is compared with the training samples by using Euclidean distance. The experimental results show that the time of one attendance is 5-10 seconds and the success rate of face recognition is 91.1%, which verifies the advantage of the algorithm in improving the efficiency of attendance.

The train wheel rim thickness is always manually measured by using handheld devices or in a non-contact way by using laser and image techniques. To reduce the operating costs, this work studied some soft measurement methods with classical machine learning algorithms including neural network (NN), locally weighted linear regression (LWLR), and support vector machine (SVM). By analyzing the correlation between features, it can improve the efficiency of the model. Some approaches were used to optimize the number of hidden layer neurons in NN, parameters in LWLR and SVM. Experiments on real data are conducted, and the results show that the proposed methods can ensure high precision and accuracy, while NN has the highest accuracy.

The current TFIDF (Term Frequency and Inverted Document Frequency) algorithm cannot effectively reflect the relationship between the importance of a word and its distribution. This paper proposes a Class Variance-Term Frequency and Inverted Document Frequency algorithm. This algorithm improves the TFIDF algorithm based on three distribution factors: category, inter-class and variance. In order to measure the optimization effect of this method, three algorithms were compared using the original algorithm, improved algorithm and TFIDF algorithm based on dual parallel calculation model. Experiments show that the improved algorithm has significantly improved recall, accuracy, and F metric values, comparing with the original algorithm, and has improved compared with the TFIDF algorithm based on dual parallel calculation model. Therefore, the improved algorithm can well adapt to the demand for feature word extraction and has better text classification performance.

The traditional dark channel priori has been successfully applied to the single image deblurring problem. According to the characteristics of the dark channel priori, clear images are recovered. However, when the fuzzy image has significant noise pollution, the dark channel prior can not play a role in the fuzzy kernel estimation. The new image denoising method based on the rational order differential inherits the advantages of the total variation denoising method which greatly improves the high frequency part of the image and the fractional order differential denoising method which can well retain the texture details of the image. In this paper, the theory of rational order differential calculation is combined with the dark channel priori of fuzzy image, and an image deblurring method based on the improved dark channel priori is proposed. The specific work is as follows: combining the maximum posterior estimation algorithm and the rational order dark channel prior, a fuzzy image model is constructed; furthermore, the model is solved by using the semi quadratic splitting method. Finally, the multi-scale iterative framework is used to estimate the fuzzy kernel of the accurate image, and then the new non blind image deblurring algorithm can be used to solve the clear image. Experimental results show that the method is effective.

The algorithm based on kernel correlation filtering shows excellent performance. Aiming at the problem that traditional correlation filtering algorithms are sensitive to deformation and cannot solve the tracking failure caused by scale changes, this paper proposes a scale-adaptive correlation filtering algorithm combining statistical color feature. Firstly, predict and train the HOG features of the correlation filter and the color histogram features to obtain the respective response values, namely the template response and the histogram response. Then adopt the block method, and perform scale prediction based on the position of each response point. Finally, the final fusion response is calculated. The experimental results demonstrate that compared with traditional KCF, when the target scale has large scale change and deformation, the improved algorithm in this paper performs better, and can still track stably.

Moving target tracking is one of the core issues in the fields of automobile assisted driving and autonomous driving. This algorithm proposes a target tracking algorithm suitable for fisheye cameras in vehicle environment. A background point removal method based on median ordering is proposed to optimize the optical flow method to solve the problem of target scale change in fisheye camera videos. Multi-scale scaling processing for frames, tracking algorithm based on cyclic matrix structure (CSK), target tracking for multiple images after zooming, using fitting to determine the maximum corresponding value, reducing the error caused by target deformation on tracking results. The tracking test results of multiple fisheye camera videos in the vehicle environment show that the algorithm has high real-time tracking accuracy.

Target tracking algorithm based on deep learning generally adopts offline trained deep network to extract features, but with unsatisfactory separation effect of the target and the background. A target tracking algorithm based on deep optimization features is proposed to solve this problem. First, the Conv3-1 and Conv4-3 layers are selected from the offline trained convolutional neural network model VGGNet16 for basic feature extraction. Then the loss function is used to quantitatively describe the effectiveness of different filter positioning target in the basic feature layer. Finally, the selected high-efficiency filters are fused to extract deep optimization features and achieve effective tracking under the correlation filtering framework. The experimental results show that compared with other related algorithms, this algorithm can better recognize the target and present better results.

This paper proposes a shortest path planning of agent in an environment based on reinforcement learning. This method adopts the Q-learning algorithm, which has gained increasingly using in agent path planning recently. This algorithm can be fully designed to a reasonable environment model and applied to other professional fields. Yet, reinforcement learning in path planning still needs to improve the rate of convergence. The learning process takes several iterations and spends a long time to find the final path. A local optimal problem in search process deals with other explorations that provided shortest path for agent to take optimal actions. For the agent, the reward value of inclined movement is introduced to find the shortest path. The autonomous obstacle avoidance is utilized to obtain the optimal path in the process. Finally, the model of intelligent agent movement environment has established in matlab. The effectiveness of the algorithm has proved by simulation results.

Focusing on the issue that the kernel correlation filter tracking method can not accurately adapt to the target scale change, a scale estimation method is added under the kernel correlation filter tracking frame, and a scale adaptive target tracking algorithm is proposed. The algorithm firstly uses the kernel function to obtain the kernel correlation filter, and estimates the position of the target for the candidate sample detection; then uses the scale estimation method to detect the target scale to further determine the optimal scale of the target to be tracked in the current frame; Finally, according to the occlusion judgment mechanism, the parameters of the target performance model and the scale model are updated online. The experimental results show that compared with kernel correlation filtering algorithm, the proposed algorithm can accurately target tracking process to adapt to changes in the target scale, and the center position error reduced by an average of 4.39 pixels.

For network control systems with long delay and packet loss, firstly, it is based on the effectiveness of model-based control in network systems with long delay, and it also establish a feedback model that simulates delay as a system component. Considering the possible modeling errors of the feedback model, a design of the traditional linear system is proposed to design the controller by using the communication interference observer to compensate for the delay characteristics. Then, the steady-state error of the characteristic equation of the system is obtained and prove the feasibility and robustness of the method, and carried out the equivalent analysis of packet loss. Finally, the effectiveness of the controller is demonstrated by the simulation of communication interference data from comparing with the Smith predictor.