Multi-level attention model for tracking and segmentation of objects under complex occlusion

Abstract

A multi-level attention framework for tracking and segmentation of humans and objects under complex occlusions is investigated, featuring an effective probabilistic appearance-based technique for pixel reclassification during object grouping and splitting. A novel 'spatial-depth affinity metric' is introduced in the conventional likelihood function, utilising information of both spatial locations of pixels and dynamic depth ordering of the component objects in grouping. Depth ordering estimation is achieved through a combination of top-down and bottom-up approach. Experiments on some realworld difficult scenarios of low quality and highly compressed videos demonstrate the very promising results achieved.

Novel person detection and suspicious activity recognition using enhanced YOLOv5 and motion feature map

“…These problems have only been partially solved. According to (Xu and Puig 2006), directional gradient histogram (HOG) is the most popular technique for human analysis (Dalal and Triggs 2005). This technique uses directional gradients to identify objects in an image.…”

Novel Pedestrian Detection and Suspicious Activity Recognition Using Enhanced YOLOv5 and Motion Feature Map

“…The most commonly utilized hand-crafted features include HOG [27], LBP [20], Integral channel features [21], Gray level co-occurrence matrix [21], CNN features [17], HAAR-Wavelet [28], and Oriented gradients [27]. As a result, the researcher turns to region-based deep learning methodologies to resolve the problems with the Boosted cascade [14] current handmade feature-based system [22], as illustrated in Table 2.…”

“…These problems are only partially resolved. The Histogram of Oriented Gradients (HOG) is the most popular technique for detecting pedestrians, according to [21]. The directional gradient is employed in this technique to identify objects in the image.…”

Group Segmentation During Object Tracking Using Optical Flow Discontinuities

“…Rowe et al [6] have proposed using a robust mean-shift method and multiple appearance models to track objects through the occlusion, rather than attempt to segment the occlusion. Xu et al [11] propose a probabilistic technique to classify pixels within an occlusion region using a combination of pixel colour and depth ordering. Denman et al [3] used colour and optical flow to locate objects and segment them during occlusions.…”

“…These techniques [6,11,10,7] all rely on colour features and function well when people are easily distinguishable and scene conditions such as lighting are stable. However when people are dressed in a similar manner as often occurs in an urban setting (i.e.…”