Gait Recognition Research Based On Ant Colony Algorithm And Genetic Algorithm

In computer vision and intelligent video surveillance system, gait recognition is a new research direction, which is to identify people based on the way people walk. Gait analysis and identification, which has important potential future applications in fields such as security, machine interaction, computer animation, virtual reality and medicine, is of great economic value. With the development of computer and information technology, gait identification technology by computer has made great progress. In order to increase the accuracy of gait recognition result, most researchers focused on the research of extracting gait feature and neglected the research on recognition algorithms. In this paper, based on the National Natural Science Foundation-funded projects, we mainly study the impact of a variety of gait recognition algorithms to identification performance.First, the paper gives a brief overview of gait recognition technology. It reviews several popular gait recognition algorithms, discusses the application of gait recognition theory and method, and summarizes the existing recognition technology. Then, the paper introduces gait feature extraction technology based on the discussion on motion detection, interesting region extraction and processing, gait cycle detection and gait energy images. For motion detection in dynamic scenes, binary motion images are obtained with mixture Gaussian background modeling. For interesting region extraction and processing, traversing search method is used to find the smallest human rectangular box. Then the human rectangular box is regularized and centered. In gait cycle detection, we get the gait cycle by calculating the auto-correlation function of the gait signal. Gait energy images, which are constructed from gait sequences, are used as classification features.The core content of this paper is to do research in a variety of recognition technologies and compare the performance of them: A gait recognition algorithm based on HMM is proposed. Baum-Welch algorithm is used to model the gait character of every body; Then Forward-Backward algorithm is used to recognize gait character of different bodies. The correct recognition rate of the algorithm based on HMM is more than 75%. It is the first time that ant colony algorithm is used in gait recognition. The algorithm simulates the adaptive process that ants find food, and can find the cluster center automatically based on the training of samples. The correct recognition rate of the algorithm enhances five percent than that of the algorithm based on HMM. We also use genetic algorithm to optimize the clustering center of ant colony algorithm. Every ant is coded as a chromosome. Then we choose or lose ants based on the fitness degree of the chromosome, and calculate the crossover probability and mutation probability. Experimental results show that the correct recognition rate of the method is more than 90%.Finally, with a large amount of experiments on CASIA database, we compare the recognition results of these three approaches and summarize the strengths and weaknesses of them on effectiveness and computational complexity. We find that the proper recognition algorithm can improve recognition result obviously and the algorithms proposed in the paper have high capability of anti-jamming which is lacked by general gait identification systems. Experiments show that the new proposed algorithms, which can enhance the quality of gait image efficiently, improve the recognition results and run faster, meet our expectation. The work of this paper provides an important reference for how to choose suitable technologies to achieve the best recognition results.