Intelligent Surveillance Video Scene Analysis technology Platform
20130306 - 20141005:
Mr Felix Chow
There will be five deliverables in this project: 1 ) Video Enhancement (requires 12 months) ASTRI will develop a state-of-the-art video enhancement technology with the Chinese University of Hong Kong. This technology can remove camera noise and compression noise, and enhance intensity range simultaneously. It is required to develop this technology because qualities of most of the surveillance videos are so low that existing algorithms fail to work. With this technology, we can improve the image quality of the videos such that these low quality videos can be processed. 2) Advanced Feature Extraction (requires 15 months) ASTRI will develop the following two new feature extraction methods for video object decomposition and classification tasks: i) Real-time and high accuracy human and vehicle detection technique; and ii) High accuracy face detection and recognition technique. These two new techniques will largely increase the respective object detection rate, which will help improve the quality of the following learning-based event detection. 3) Learning-based Event Detection (requires 11 months) ASTRI will develop an automatic behavior learning technique to learn and understand the motion behavior of moving objects captured by a video camera. It will generate a classifier for normal-and-abnormal motion classification. The classifier can be used to detect abnormal events for the learned camera by automatically generating rules to analyze the moving content. This kind of technique is not supported in existing intelligent video surveillance products. 4) Query-by-Example/Sketch Object Indexing and Retrieval (requires 5 months) ASTRI will develop an object retrieval technique that allows a user to input images or sketches as queries for searching the desired objects from an analyzed video. This kind of technique is not supported in existing intelligent video surveillance products. 5) An integrated client-server system (requires 4 months) ASTRI will integrate all the above technologies together with the necessary low-level video processing techniques from previous projects into a client-server system, which provides all the necessary services for video surveillance analysis. Some of the above deliverables will be developed concurrently. The total project duration is 19 months. Please refer to Annex 1: Project Schedule for the detailed timeframe.
Dr Shen-Chang Chao Dr I Sheng Tang Dr Tai-Pang Wu Dr Lawrence Chun-Man Mak Ms Xiao Zhou Mr Yee-Lim Chan Mr Hangiang Huang Dr Jacob Peijie Huang Mr Hongzhi Guo Mr Chi Wai Lam Mr Hui Li Ms Jianting He Dr I Keong Albert Lau Ms Hoi Sim Wong Mr Hon Bill Yu Mr Chiu Wa Ng Mr Joni LAITINEN Ms Shan Shan Zheng Mr Qiang Liu Ms Xinghua Zhu
Amplesky communication technology (Beijing) Co. Ltd. [Sponsor] Guilin Li Da communication engineering Co.Ltd HK Correctional Services Department (CSD) Hong Kong Jiuling Technology Co. Ltd [Sponsor] Hong Kong Jiuling Technology Co. Ltd (contract service) [Sponsor] The Chinese University of Hong Kong
In recent years, the demand of Surveillance Video Analysis Platform, which drives the expansion of the respective market size, is rapidly increasing. As a consequence of the increasing demand, there exists a huge amont of surveillance videos that are waiting to be processed. While existing intelligent surveillance analysis systems generally focus on low-level algorithms such as video object decomposition and classification, this approach imposes difficulty on the system for analyzing the content of the input video because such approach requires a user to specify manually a set of rules for different scenarios and different videos by using the low-level algorithms. It is impractical for such a system to process a huge amount of videos. The main objective of this project is to develop a surveillance video analysis system that is able to truly ‘analyze’ and ‘understand’ the content of surveillance video. The system will learn and understand the motion behavior of the moving objects such that manual operation of a user will no longer be required. This system will fill the respective market gap. To facilitate the use of the analyzed content, the second objective of the project is to develop a user-friendly ‘query-by-example’ video object retrieval system. To achieve the two ultimate goals, it is necessary to modify and enhance existing video enhancement, object decomposition and classification techniques. Part of the mentioned techniques will be transferred to front-end device to leverage the intelligence of the devices. The whole system will be deployed as a client-server system. ASTRI will work closely with Hong Kong Disciplined Forces, including Hong Kong Correctional Services Department, and industrial partners to develop a cutting-edge Surveillance Video Analysis Platform.