CN104281837B - With reference to Kalman filtering and the adjacent widened pedestrian tracting methods of interframe ROI - Google Patents

Abstract

The present invention proposes a kind of method that Kalman filtering is combined progress pedestrian tracking with adjacent interframe expansion area-of-interest method.The present invention provides a kind of combination Kalman filtering and the adjacent widened pedestrian tracting methods of interframe ROI, the method overcome pedestrian tracking treatment effect is undesirable on occlusion issue the problem of, is very suitable for the real-time tracking to pedestrian.The experimental results showed that this method can be predicted the position of current pedestrian using tracing positional information before, improve the accuracy rate of pedestrian tracking well when pedestrian is blocked.

Description

Pedestrian Tracking Method Combining Kalman Filter and ROI Expansion Between Adjacent Frames

technical field

The invention belongs to the technical field of video image processing and pattern recognition, and in particular relates to a method for tracking pedestrians by combining Kalman filtering with a pedestrian ROI expansion method between adjacent frames.

Background technique

Pedestrian tracking is a hot issue in non-rigid moving target tracking in the field of machine vision. As a frontier science in the field of computer vision, it involves image processing, pattern recognition and artificial intelligence. It is a pedestrian target in image sequences or video surveillance. Detecting, extracting, identifying, and tracking, and obtaining parameters such as the position, velocity, acceleration, and trajectory of pedestrians are important steps to analyze pedestrian behavior and gain a deeper understanding of behavior.

This problem is divided into two parts: pedestrian detection and pedestrian tracking. Pedestrian detection belongs to moving target detection. The purpose is to extract the changing area (pedestrian) from the background image from the sequence image. According to different tracking methods, pedestrian tracking is divided into four categories: model-based tracking, area-based tracking, active contour-based tracking and feature-based tracking. Due to the non-rigid characteristics of pedestrians and the flexibility and autonomy of pedestrian activities, one difficulty in specific applications is the occlusion problem. For the occlusion problem, the above pedestrian tracking methods are not ideal.

Contents of the invention

In order to solve the problems existing in the prior art, the present invention proposes a pedestrian tracking method that combines Kalman filtering with a pedestrian region of interest (ROI) expansion method between adjacent frames to improve pedestrian tracking accuracy.

In order to achieve the above object, the present invention has taken the following technical solutions:

A pedestrian tracking method combining Kalman filtering and pedestrian ROI expansion between adjacent frames, characterized in that the method comprises the following steps:

Step A: The previous frame image is detected by the single frame image method, the coordinates of the detected rectangular frame are extracted and saved, and the pedestrians in the current frame image are tracked and identified according to the detection result of the previous frame image, and the detection result is output;

Step B: According to the detection result of step A, judge whether there is a detection target, if there is no detection target, execute step C, otherwise, execute step D;

Step C: According to the recording results of the previous frames in the history tracking module, use Kalman filter to predict the current frame image, and obtain the prediction result to solve the tracking problem of pedestrians being blocked for a short time, and then perform step D;

Step D: mark the position of the pedestrian according to the detection result/prediction result, save the position of the pedestrian in the history tracking module and output the position of the pedestrian.

Further, said step A includes the following steps:

Step A1: Collect the image in front of the vehicle and perform scan window matching detection on the image, extract and save the detection result;

Step A2: Change the horizontal and vertical coordinates of the rectangular frame to enlarge it to a certain ratio as the region of interest, detect it as the detection region of the next frame, and judge whether the window will increase according to the lower boundary area of the scanning window in the detection region.

Further, after said step A2, step A3 is also included: in the process of detecting between adjacent frames, it is judged whether a full-scan image is needed according to the frequency of pedestrian update, if necessary, the image in front of the vehicle is re-acquired, and if not, continue to extract The detection result of step A1 is used as the ROI of the next frame.

The beneficial effects of the present invention are: the present invention overcomes the shortcomings of pedestrian tracking algorithms in the prior art, and is very suitable for real-time tracking of pedestrians. It not only uses the image information between adjacent frames to improve the calculation speed, but also combines the Kalman filter Come in to make up for the lack of frame processing that cannot track pedestrians being blocked for a short time. Experimental results show that the method of the present invention can use the previous tracking position information to predict the current position of the pedestrian when the pedestrian is blocked, and improve the accuracy of pedestrian tracking.

Description of drawings

Fig. 1 is the flowchart of the ROI expansion method between adjacent frames adopted by the method of the present invention;

FIG. 2 is a flow chart of the pedestrian tracking method combined with Kalman filtering and ROI expansion between adjacent frames of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

The target in the moving area in the single frame image is detected and recognized, and whether the target is a pedestrian is confirmed according to the extracted features to determine whether there is a pedestrian target in the image sequence, and then the coordinate position of the pedestrian is obtained according to the detection result. In general, after the pedestrian target is confirmed, it will not disappear immediately, and it exists continuously in the sequence images. That is, the pedestrian's position has a certain continuity, and its characteristics also have a great correlation. For this reason, the present invention adopts the method of establishing a region of interest (ROI) for multi-frame images, and uses a single-frame image detection method to find pedestrians in the region of interest, which can effectively reduce tracking time. Considering that the height of pedestrians will not change greatly during walking, but the width of pedestrians will change significantly due to changes in pedestrian walking gait. In this way, in the recognition result of the previous frame, a certain area can be expanded in the height direction and width direction of the pedestrian rectangular frame respectively as the area of interest, and then the pedestrian in the current image can be accurately located in this area. The process of pedestrian detection between adjacent frames not only greatly improves the detection speed of single frame images, but also optimizes the images between adjacent frames.

Accompanying drawing 1 is the flowchart of pedestrian detection process between adjacent frames. First, collect the image in front of the vehicle and perform scan window matching detection on the image, extract the detection result and save it; then change the horizontal and vertical coordinates of the rectangular box to enlarge it to a certain ratio as the region of interest, and use the region of interest as the next frame The detection area is detected and whether the window will increase can be judged according to the lower boundary area of the scanning window in the detection area. In the process of detection between adjacent frames, it can be judged according to the update frequency of pedestrians whether a full-scan image is needed, if necessary, the image in front of the vehicle will be collected, and if not, the detection result will continue to be extracted as the ROI of the next frame.

Accompanying drawing 2 is the flow chart that combines Kalman filtering and the pedestrian ROI expansion method between adjacent frames for pedestrian tracking. A typical example of Kalman filtering is to predict the coordinates and velocity of the object's position from a limited set of noise-containing observation sequences (possibly biased) about the object's position. The present invention uses Kalman filter to deal with the tracking problem when pedestrians are blocked, and combines Kalman filter with the method of enlarging the scanning window as the region of interest to achieve more accurate tracking of pedestrian targets and shorten the detection time. The method of the invention closely links the picture information between adjacent frames, and greatly reduces the complexity of system processing.

First, the pedestrian detection module gives the target position Rn-1 detected by the previous frame image, expands the original scanning rectangular frame by a certain multiple as the region of interest, and detects the region to obtain the detection result An, and then judges the Whether there is a detection target in the previous frame in the detection result, if there is a target (e=1), mark the pedestrian position in the current frame picture and save it in the history tracking module for the detection of the next frame, that is, Rn=An; If there is no target (e=0), the present invention proposes a method for anti-occlusion prediction using Kalman filter, according to the record results of the previous frames in the history tracking module, using Kalman filter to predict the current frame image, and save it in the history Tracking module, ie Rn=Kn. Among them, Rn is the record in the history tracking module, An is the detection result of ROI between adjacent frames, and Kn is the prediction result of Kalman filter anti-occlusion.

The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be assumed that the specific implementation of the present invention is limited to these descriptions. For those of ordinary skill in the technical field of the present invention, without departing from the concept of the present invention, some simple deduction or replacement can be made, which should be regarded as belonging to the protection scope of the present invention.

Claims (1)

1. a kind of pedestrian tracking method that combines Kalman filtering and the pedestrian ROI expansion between adjacent frames, it is characterized in that, described method comprises the following steps: Step A: The previous frame image is detected by the single frame image method, the coordinates of the detected rectangular frame are extracted and saved, and the pedestrians in the current frame image are tracked and identified according to the detection result of the previous frame image, and the detection result is output; Specifically, the step A includes the following steps: Step A1: Collect the image in front of the vehicle and perform scan window matching detection on the image, extract and save the detection result; Step A2: By changing the horizontal and vertical coordinates of the rectangular frame to enlarge it to a certain ratio as the region of interest, detect it as the detection region of the next frame and judge whether the window will increase according to the lower boundary region of the scanning window in the detection region; A3: In the process of detecting between adjacent frames, judge whether a full-scan image is needed according to the update frequency of pedestrians. If necessary, re-acquire the image in front of the vehicle. If not, continue to extract the detection result of step A1 as the interest of the next frame. area; Step B: According to the detection result of step A, judge whether there is a detection target, if there is no detection target, execute step C, otherwise, execute step D; Step C: According to the recording results of the previous frames in the history tracking module, use Kalman filter to predict the current frame image, and obtain the prediction result to solve the tracking problem of pedestrians being blocked for a short time, and then perform step D; Step D: mark the position of the pedestrian, save the position of the pedestrian in the history tracking module and output the position of the pedestrian.