Search

Objective perspective distortion is a problem that needs to be solved by video surveillance analysis. Compared with the street scene method, which depends on prior knowledge of the scene or 3D scen...

Street scene change detection continues to capture researchers' interests in the computer vision community. It aims to identify the changed regions of the paired street-view images captured at different times. The state-of-the-art network based on the encoder-decoder architecture leverages the feature maps at the corresponding level between two channels to gain sufficient information of changes. Still, the efficiency of feature extraction, feature correlation calculation, even the whole network requires further improvement. This paper proposes the temporal attention and explores the impact of the dependency-scope size of temporal attention on the performance of change detection. In addition, based on the Temporal Attention Module (TAM), we introduce a more efficient and light-weight version - Dynamic Receptive Temporal Attention Module (DRTAM) and propose the Concurrent Horizontal and Vertical Attention (CHVA) to improve the accuracy of the network on specific challenging entities. On street scene datasets `GSV', `TSUNAMI' and `VL-CMU-CD', our approach gains excellent performance, establishing new state-of-the-art scores without bells and whistles, while maintaining high efficiency applicable in autonomous vehicles., 8 pages, 9 figures, 6 tables. Accepted to IEEE Intelligent Vehicles Symposium 2021 (IV2021). Code is available at https://github.com/Herrccc/DR-TANet

It starts with a dress, or dresses. Among a menagerie of rainbow variations, certain features are standard: lace and ruffle-decked blouses under jumpers, aprons, or high-waisted belled and crinolined knee-length skirts; more skin covered than bare; headwear including bonnets, miniature hats, or massive bows over ringlets and long tresses. So many ruffles; so much lace (Fig. 1). Beginning in the late 1990s,gothic lolitas—overwhelmingly young women in their teens and twenties, and overwhelmingly girly in their outsized bows, platform Mary Jane shoes, and petticoated skirts—stood out as defiantly, bizarrely out of place and time on the Tokyo street scene, all bright white and concrete in Harajuku, a built-up postwar neighborhood of Tokyo known as a youth haven since the 1960s. More than twenty years later, although most Harajuku fashions have died out in keeping with a fad's typically short life cycle, the gothic lolitas have persisted and even multiplied, thanks in large part to the Internet, which has helped muster an army of misfit girl aristocrats not just in Japan but around the globe.

SegNet is a Convolution Neural Network (CNN) architecture consisting of encoder and decoder for pixel-wise classification of input images. It was found to give better results than state of the art pixel-wise segmentation of images. In proposed work, a compressed version of SegNet has been developed using Differential Evolution for segmenting the diseased regions in leaf images. The compressed model has been evaluated on publicly available street scene images and potato late blight leaf images from PlantVillage dataset. Using the proposed method a compression of 25x times is achieved on original SegNet and inference time is reduced by 1.675x times without loss in mean IOU accuracy.

Driving and biking are complex and attention-demanding tasks for which distractions are a major safety hazard.Modeling driving-related attention with regard to audio-visual distractions and assessing the attentional overload could help drivers to reduce stress and increase safety. In this work, we present a multimodal recording architecture using dry EEG-electrodes and the eye-tracking capability of the HoloLens 2 for an outdoor Augmented Reality (AR) scenario. The AR street scene contains visual distractions and moving cars and is shown to the subject in a simulation through the Hololens 2. The system records EEG and eye-tracking data to predict changes in the driver's attention. A preliminary case study is presented here to detail the data acquisition setup and to detect the occurrences of visual distractions in the simulation. Our first results suggest that this approach may overall be viable. However, further research is still required to refine our setup and models, as well as to evaluate the ability of the system to capture meaningful changes of attention in the field.

Locale is one of the basic elements of place, referring to the physical settings and visual appearance of a place. Understanding and representing a locale is of great importance in terms of human perception and human activity. However, taking a quantitative measurement of the visual appearance of urban environment has proven to be challenging because visual information is inherently ambiguous and semantically impoverished. To mitigate this issue, this paper employs street-level images as the proxy for urban physical appearance, utilizes the recently developed image semantic segmentation techniques to parse an urban scene into scene elements, and proposes a framework for locale representation using scene elements. The framework is composed of two major components: street scene ontology and street visual descriptor, which are aimed at street scene qualitative understanding and quantitative representation respectively. A case study is developed to demonstrate the application and advantage of the street scene ontology and street visual descriptor. A series of quantitative analyses demonstrates the ability and great potential of the framework for investigating the connections between place and other socioeconomic factors.

Fear of crime, which may be present without experiencing an actual crime, can restrict one’s daily physical and mental activities and reduce quality of life. In previous research, fear of crime was measured by regional surveys. Though useful for confirming group characteristics, regional surveys cannot measure in real-time, assess individual characteristics, or provide an objective measure of anxiety. Since the causes and effects of fear of crime are highly individualized, we have developed a protocol to measure physiological signals in concert with existing surveys; this system can verify an individual’s fear of crime characteristics in real time. Subjects were shown 6 clips of actual pedestrian environments (day/night of a commercial street scene, day/night of a residential street scene, day/night of a natural street scene). To ease immersion of the subjects into the scenes, clips were produced from the subjects’ first-person point of view. Subjects were divided into two groups (Group1: N = 14, age, 22 ± 1.66 years; Group2: N = 13, age 21 ± 1.35 years) based on their fear intensity as reported on our pre-recording survey; electroencephalographic (EEG), electrocardiographic (ECG), and galvanic skin response (GSR) signals were compared between groups. They were then assessed via video for comparative purposes. Our results demonstrated that the physiological signals were dependent on how conscious an individual was of his or her own fear of crime. We found significant differences between the two groups for all video clips except for daytime commercial street and nighttime natural street; these data suggest that individual characteristics are important in measuring fear of crime.