We have introduced a new part-based model for identifying twelve body parts from the extracted human silhouettes. According to this approach, the segmented silhouettes are first converted into binary silhouettes and then their contours are obtained. Further, convex hulls are drawn around the contours. Points on the convex hull that are also part of the original contours are thus identified. Then, only five such points are chosen since having more than one point on the same body part is useless. In cases, where multiple points are detected on the same body part, only one is selected through a point elimination technique based on Euclidean distance. Furthermore, a sixth point is obtained by finding the centroid of the contour. Using the obtained six points, six additional key points are then extracted. The mid-point of two key points is found and a point on the contour lying closest to the obtained mid-point is stored as an additional key point. For example, the right elbow lies between the head and the right hand. Then the Euclidean distance between this midpoint and all other points lying on the human contour are calculated and the point having the minimum distance is selected as the right elbow point. Similarly, the hip joints are found by calculating the midpoints of the head and the feet. Likewise, the knee joints are identified by finding the midpoints of the torso and the feet. Each step of the process is shown in this figure:

Waheed, M., Jalal, A., Alarfaj, M., Ghadi, Y. Y., Al Shloul, T., Kamal, S., & Kim, D. S. (2021). An LSTM-Based Approach for Understanding Human Interactions Using Hybrid Feature Descriptors Over Depth Sensors. IEEE Access, 9, 167434-167446.
Waheed, M., Javeed, M., & Jalal, A. (2021, November). A Novel Deep Learning Model for Understanding Two-Person Interactions Using Depth Sensors. In 2021 International Conference on Innovative Computing (ICIC) (pp. 1-8). IEEE.

Human Activity Recognition (HAR), Human Interaction Recognition (HIR), Human-Object Interaction Recognition (HOIR), Human Motion Tracking.

The code for our paper is publicly available for use in academic and research activities. Please feel free to drop an email to Dr. Ahmad Jalal ([email protected]) or visit our website: https://portals.au.edu.pk/imc Download Code: Automated Part-based Model

We build a new online continuous interaction dataset (IM-IntensityInteractive7) with depth sensor. The dataset is designed to cover human daily interactions collected in an indoor environmental setting. There are seven types of interactions: handshake, fight, push, punch, greet, kick and hug. Most of the interactions in the dataset are highly similar to each other, which make this dataset quite challenging. Dataset includes 12 segmented videos sequences of each human interaction for training and 8 un-segmented continuous videos sequences for testing.
For training the system we used 9 subjects who performed different human interactions in pairs.
In the testing sets, we used 8 un-segmented continuous interaction videos sequences from 4 new subjects. Each subject performed 4 un-segmented videos sequences.
Format: .avi, .jpeg files
Camera: Bumble Bee Camera

A. Jalal, M. Mahmood and A. S. Hasan "Multi-features descriptors for human activity tracking and recognition in Indoor-outdoor environments," IEEE International Conference on Applied Sciences and Technology, 2019.
A. Jalal and M. Mahmood, “Students’ Behavior Mining in E-learning Environment Using Cognitive Processes with Information Technologies,” Education and Information Technologies, Springer, 2019.

Human Interaction Recognition, Human to Human Relation, Human Motion Tracking.

This dataset is freely available to academic or research activities. Please feel free to drop an email to Dr. Ahmad Jalal ([email protected]) or visit portals.au.edu.pk/imc Download code for these applications kernel sliding perceptron

We have introduced a Sporting Behaviors dataset (IM-SportingBehaviors) using triaxial accelerometers attached to the subject's wrist, knee, and below neck region to capture important aspects of human motion. The dataset represents motion data captured while subjects are involved in performing 6 sporting behaviors: badminton, basketball, cycling, football, skipping, and table tennis. The subjects involved both professional and amateur athletes who aged between 20-30 with having weight ranging between 60-100 kgs. Due to multi-sensor environment and inter-behavioral similarity, the dataset itself poses adequate amount of challenges.
- Dataset includes 120 sequences of acceleration data from three body worn accelerometers with variable time duration (between 40s to 60s)
- For training the system, we used 20 subjects who performed behaviors in repetitive nature
- In testing sets, we used a combination of repetitive and passive movement from each behavior set
Format: .txt files
Naming: [GENDER]-[VOLUNTEER]-[DATED]

• [GENDER]: 'm' for males, and 'f' for female
• [VOLUNTEER]: represents the volunteer number
• [DATED]: Date is formatted as [YYYY-MM-DD-HH-MM-SS]

A. Jalal, Majid A. K. Quaid, and A. S. Hasan, “Wearable Sensor-Based Human Behavior Understanding and Recognition in Daily Life for Smart Environments," IEEE International Conference on Frontiers of information technology, 2018.
A. Jalal, M. A. K. Quaid and M. A. Sidduqi "A Triaxial acceleration-based human motion detection for ambient smart home system," IEEE International Conference on Applied Sciences and Technology, 2019.

Human Behavior Recognition, Human Movement Tracking.

This dataset is freely available to academic or research activities. Please feel free to drop an email to Dr. Ahmad Jalal ([email protected]) or visit portals.au.edu.pk/imc Download Dataset IM-SportingBehaviors Dataset Download code for these applications RGA

We have introduced a smart home dataset using three triaxial IMU(inertial measurement unit) sensors attached to the subject wrist, chest, and thigh region to capture important aspects of human motion. The dataset represents motion data captured while subjects are involved in performing 11 different (static and dynamic) smart home activities: Using Computer (1min), phone conversation (1 min), vacuum cleaning (1min), reading book (1 min), watching tv(1 min) , ironing (1 min), walking (1 min), exercise (1 min), cooking (1 min), drinking (20 times), brushing hair (20 times). The subjects involved both young and old volunteers who aged between 19-60 with having weight ranging between 55-85 kgs. Due to multi-sensor environment, most of the movements in activities are highly similar to each other, which make this dataset quite challenging.
- Dataset includes 220 sequences of inertial data from three body worn inertial measurement unit sensors with variable time duration (between 45s to 60s)
- For training the system, we used 10 subjects who performed activities in repetitive in nature.
- In testing sets, we used a combination of repetitive and passive movement from each activity set.
Format: .txt files
Naming: [Gender]: ‘f’ for female, and ‘m’ for male.

Tahir, S.B.; Jalal, A.; Kim, K. “Wearable Inertial Sensors for Daily Activity Analysis Based on Adam Optimization and the Maximum Entropy Markov Model”, Entropy, 2020.
Tahir, S.B.; Jalal, A.; Batool, “M. Wearable Sensors for Activity Analysis Using SMO-based Random Forest over Smart home and Sports Datasets”, IEEE International Conference on Advancements in Computational Sciences,2020.

Wearable Sensors, Healthcare monitoring, Smart Homes, Human Activity Recognition.

This dataset is freely available to academic or research activities. Please feel free to drop an email to Dr. Ahmad Jalal ([email protected]) or visit portals.au.edu.pk/imc Download Dataset IM-Wearable Smart Home Activities (IM-WSHA) Dataset

This dataset is freely available for academic or research activities. Please feel free to drop an email to Dr. Ahmad Jalal ([email protected]) or visit portals.au.edu.pk/imc Download Dataset ieSense: Sensor Based Real Time Pervasive Indoor Position and Environment-Type Detection