Snoopertrack: Text detection and tracking for outdoor videos

Abstract

In this work we introduced SnooperTrack, an algorithm for the automatic detection and tracking of text objects -such as store names, traffic signs, license plates, and advertisements -in videos of outdoor scenes. The purpose is to improve the performances of text detection process in still images by taking advantage of the temporal coherence in videos. We first propose an efficient tracking algorithm using particle filtering framework with original region descriptors. The second contribution is our strategy to merge tracked regions and new detections. We also propose an improved version of our previously published text detection algorithm in still images. Tests indicate that SnooperTrack is fast, robust, enable false positive suppression, and achieved great performances in complex videos of outdoor scenes.

Recognition of Video Text through Temporal Integration

“…Thus, it is more robust to background changes than methods that do not attempt to identify the text pixels. For example, [2], [6], [7] perform tracking by extracting features from all pixels in a bounding box, including background pixels. Moreover, SWT is rotationinvariant, and SIFT is robust to rotation, scale change and viewpoint change.…”

“…Thus, two instances of the same text, but with drastically different backgrounds, may be wrongly classified as two different text objects. In the second category, many features have been explored for text tracking: difference of intensity values [5], horizontal and vertical projection profiles of gradient magnitudes [2], motion vectors in the P-frames of MPEG videos [6] and Histogram of Oriented Gradients [7]. However, because these features are not robust to rotation and viewpoint change, it is difficult to extend them to handle complex text movements.…”

T-HOG: An effective gradient-based descriptor for single line text regions

“…The motivating application for text classifiers such as T-HOG and R-HOG is the detection of text in photos and videos of arbitrary scenes [29,30]. Specifically, the idea is to use the classifier to filter the output of a fast but "permissive" (high-recall and moderate-precision) detector.…”

MSER-Based Real-Time Text Detection and Tracking

“…This yields an extra speed-up that can be exploited in see-though applications (e.g. Augmented Reality translation [6], [7] and augmented documents [9]) or street-view navigation [5].…”

“…Camera based scene text analysis applications targeted specifically for mobile and wearable devices is an interesting area of research receiving increasing attention [1], [2], [3], [4], [5], [6], [7]. Although the newly arrived products in the mobile device market (2013) feature high definition cameras of up to 12 mega-pixel sensors, and powerful quad-core processors, they still have many limitations in comparison with standard desktop computers: e.g.…”