Deep-learning based single object tracker for night surveillance

Abstract

Tracking an object in night surveillance video is a challenging task as the quality of the captured image is normally poor with low brightness and contrast. The task becomes harder for a small object as fewer features are apparent. Traditional approach is based on improving the image quality before tracking is performed. In this paper, a single object tracking algorithm based on deep-learning approach is proposed to exploit its outstanding capability of modelling object’s appearance even during night. The algorithm uses pre-trained convolutional neural networks coupled with fully connected layers, which are trained online during the tracking so that it is able to cater for appearance changes as the object moves around. Various learning hyperparameters for the optimization function, learning rate and ratio of training samples are tested to find optimal setup for tracking in night scenarios. Fourteen night surveillance videos are collected for validation purpose, which are captured from three viewing angles. The results show that the best accuracy is obtained by using Adam optimizer with learning rate of 0.00075 and sampling ratio of 2:1 for positive and negative training data. This algorithm is suitable to be implemented in higher level surveillance applications such as abnormal behavioral recognition.

Classification of tomato leaf diseases using MobileNet v2

“…Training data is the dataset used to train the MobileNet V2 (weights and biases in the case of standard CNN), while testing data is the sample that is used to evaluate the performance of the trained network. Inspired by [26], four ways of data division as shown in Table 4 5 shows the classification accuracy when the batch size is increased from 16 to 48. The best performance is obtained when the batch size of 16 is used with an accuracy value of 0.9594.…”

Pedestrian age estimation based on deep learning

“…Although the black and white face images are robust in face recognition but they are excluded, The VGGFace2 dataset focused on facial and image variation due to color processing as shown in Figure 4. Five age classes have been included in this study {(00-10), (11)(12)(13)(14)(15)(16)(17)(18)(19)(20), (21)(22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35), (36-54), (55-90)}.…”

“…Before loading input images, the working environment need some preparation such as installing the required libraries, then start to load the dataset. A preprocessing data, is required at this stage including cleansing the dataset from low quality images which will confuse the training model, A relabeling age group images to five categories {(00-10), (11)(12)(13)(14)(15)(16)(17)(18)(19)(20), (21)(22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35), (36-54), (55-90)}, splitting dataset to 80% train, 10% test and 10% validation and finally, set the batch size. The age estimation model is trained with three pre-trained-weight, and two deep learning algorithms (VGG-Face and ResNet50) as it illustrated in the experimental work.…”

“…A deep CNN's mathematical construction consists of several types of layers: convolutional layers, pooling, and a connected layer. The convolution and pooling layers' function are to perform feature extraction, while the fully connected layer feds with the result of these processes to get the final classification decision [10], [11].…”

Development and performance evaluation of object and traffic light recognition model by way of deep learning

“…Jasm et al [19] implemented the image classification using convolutional neural network (CNN) on Canadian Institute for Advanced Research, 10 classes (CIFAR-10) dataset. Kadim et al [20] used pre-trained CNN alongwith fully connected layers to handle challenges for dealing with chages in appearance. Also different hyperparameters, learning rate and ratio of training samples are otimized for tracking in night conditions.…”