Automatic Detection and Tracking of Objects in Video Data with Global Motion

Abstract

The automatic method of non-point objects of interest detection and tracking in video data obtained by a video camera mounted on a mobile carrier is proposed. Additional features of the problem are a non-uniform background, the presence of objects overlapping with the background and each other, significant and rapid changes in the object of interest size of. The automatic detection is based on a convolutional neural network with YOLO architecture. Due to limitations on computing resources, object tracking is implemented without neural network solutions. To ensure stable tracking, several detectors are used simultaneously with subsequent analysis of the obtained data. The tracking stage is based on a detector based on histograms of oriented gradients (HOG), supplemented by a detector based on correlation filtering and motion trajectory prediction based on the Kalman filter. The proposed method allows detecting and successfully tracking objects at the distance of 1500 meters with an object projection size on the frame 5 x 5 pixels in conditions of global movement, non-uniform background and significant dynamics of object of interest properties. At the automatic detection stage TPR averaged over all video files participating in the experiments corresponds to 0.81, FPR corresponds to 0.10. At the tracking stage, the failure rate (tracking failures) is 6*10^(-5)

Keywords

• object detection
• image processing