Research On Pedestrian Detection Algorithm Based On Improved KSVD And Extreme Learning Machine

The illegal driving of pedestrian and non-motor vehicle will affect the normal operations of the highways, and even cause serious traffic accident. Therefore, pedestrian detection has great significance to guarantee the normal operation of highway. Pedestrian detection is a process to extract and identify the foreground target of the video. However, the illumination variation, branches swaying and the change of pedestrian posture and wear add to the difficulty of the pedestrian detection. The paper mainly researches the prospect target extraction method and the feature extraction method of the target region. The method can rapidly and accurately detect the violations of the pedestrian, the main research contents are as follows.(1) Put forward a moving object detection algorithm of the improved three frame difference algorithm combined with Gaussian mixture background difference. First of all, it improves the three frame difference algorithm combined with the edge detection algorithm, uses the Sobel operator to extract the edge information of three consecutive frames, does the difference operation on the first two frames and the last two frames, and does the 8 neighborhood summation operation on the gotten difference images. Then the method of mixed Gauss background modeling is introduced to the improved three frame difference algorithm. The experimental results show that the method has strong adaptability to the environment, and has good detection effect.(2) A multi-target tracking algorithm is proposed based on Camshift algorithm and Kalman filter. Firstly according to the detected moving region, use the Kalman filter to estimate the target position of the current frame image, and use Camshift algorithm to find target similar to the template in the neighborhood of the value predicted by Kalman filtering algorithm. Finally, use the optimal target location based on Camshift algorithm to replace the observation value of Kalman filter to update status of the Kalman filter. The experimental results show that the method can automatically calibrate target area and adapt to the target of high-speed movement, background interference and shape size changes, and so on.(3) Put forward a pedestrian detection method based on the improved KSVD and extreme learning machine. Firstly, it makes improvement on the initialization method of the dictionary for KSVD, and using the improved KSVD to get the characteristics dictionary and the corresponding sparse coefficient, then take the characteristics dictionary and the corresponding sparse coefficient into extreme learning machine for classification. The experimental results show that the improved KSVD method has improved the recognition performance of the dictionary. Under the premise of the fast recognition rate, the method has improved the accuracy of pedestrian recognition.The paper proposes a pedestrian detection method based on improved KSVD and extreme learning machine for the special application environment of expressway. It excellently completes the extraction, tracking and classification of prospect target, and has important significance to other scenarios of pedestrian detection technology.