ACTi Receives a Perfect Score in the Taipei City Police Department Intelligent Surveillance System Tender Evaluation!

Related News:
1. Mayor Hau Promotes the Full Deployment of Video Surveillance Systems and Enforces a Zero-Tolerance Public Safety Policy, Taipei City Government Spokesperson’s Office, http://jad.police.taipei/ct.asp?xItem=980802&ctNode=82918&mp=108161M

Competition Notes:
I was the engineer responsible for six intelligent video surveillance analysis algorithms in this tender evaluation, namely people and vehicle counting, unattended object detection, restricted area detection, tripwire detection, object direction detection, and object color detection. Their intended applications are briefly described as follows:
1. People and vehicle counting: can be used for traffic flow control and evaluation.
2. Unattended object detection: can detect illegal dumping or suspicious objects left behind (for example, explosives or lost property).
3. Restricted area detection: can automatically detect whether an object has entered a dangerous area.
4. Tripwire detection: can automatically detect whether an object has crossed a restricted boundary (for example, if someone is about to jump into a river).
5. Object direction detection: can automatically detect vehicles traveling in the wrong direction.
6. Object color detection: assists the police in searching for suspicious vehicles by color.

Five companies participated in this competition. I was the engineer from one of them responsible for these six algorithms; the company was ACTi Corporation. The competing systems included the IBM S3 system, the V5 systems developed by two National Chiao Tung University teams, and one independently developed system from another company. The competition was conducted using live, on-site video processing, with results recognized in real time and recorded on video for verification. Because it was a real-time competition, weather conditions had to be taken into account. Changes in light and shadow could easily cause the algorithms to fail or generate false detections. In particular, a few days before the competition, it was reported that Typhoon “Nangka” might strike Taiwan, so recognition under rainy conditions became especially important. Our team therefore continued to work in the rain on the streets. Unexpectedly, the competition day turned out to be bright and sunny. Things do not always go as planned.

The challenge of digital image analysis lies in its unpredictability. Variations are difficult to control. Simply put, if you point a camera at a white wall for ten minutes and then inspect each frame separately, the numerical values in every frame will be different. This is why mathematical analysis and abstraction are especially important.

Every company gave its all in this competition. We all arrived early at the competition site every day, constantly testing, verifying, and refining our systems. There were field engineers, core system engineers, algorithm programmers, managers, sales engineers, application engineers, sales staff, and many others. Passersby even thought we were holding a street fair, because each company had set up its own display booth by the roadside, and it almost felt as if we were living there. After all, this was a major project. Since the competition primarily evaluated algorithms, the engineers responsible for algorithm design were under tremendous pressure, because success or failure depended on the algorithms. For that reason, I often woke up in fear after falling asleep. I would sleep for less than three hours before waking up and thinking through problems again. Sometimes, when inspiration struck, I stayed up all night continuously experimenting and revising. To keep my mind clear, I ate only one meal a day and took vitamins. As the competition approached, I simply kept drinking coffee, trying to improve as much as possible. I believe engineers from the other companies must have felt the same way. The difference was that my supervisor was understanding. He only asked me to communicate clearly with my colleagues about each part of the process every day, and then he let me continue working. Although I knew he was under great pressure as well, I could not help him until the algorithms were fully ready.

Our sales engineers and application engineers continuously provided test videos. Whether the weather was cloudy, lightly rainy, or sunny, they kept recording footage so that I could train the algorithms under many different scenarios. There were many problems that had to be solved. For example, to simulate future cameras mounted on traffic lights, we had to account for camera shaking caused by wind. The algorithm had to eliminate that motion before recognition could proceed. Camera lenses could also produce a “breathing” effect, which again had to be filtered out by the algorithm. Changes in color balance also had to be handled algorithmically. In addition to these issues, there were other rule-based challenges, such as how to count accurately, how to define color judgment rules, how long an object must remain before it can truly be considered “unattended,” and how to confirm the direction of a moving vehicle. Building a practical real-time video recognition system is truly not easy. On the software side, I worked closely with a core system engineer. In addition to being highly skilled in system programming, he also raised many reasonable questions about the performance of my algorithms. I accepted those concerns and continued revising them day and night. Every time I revised my algorithms, he had to modify his code accordingly, yet he never complained. In the end, he packaged my algorithms into an ActiveX Control and turned them into a highly polished commercial-grade software product. Just by looking at the display and user interface, one could easily mistake it for a boxed retail product.

At last, the day of the competition arrived. Thanks to our prior efforts, our algorithms had become increasingly practical and were able to handle real-time changes. For example, when the weather shifted from bright sunshine to sudden cloud cover, our system could still cope. By contrast, a static background method would certainly have failed. On that day, one company requested time before each test for its system to learn the background, and the consulting panel agreed. Before the competition, the consulting panel held meetings with all participating vendors, and the evaluation proceeded only after fairness had been confirmed. Preparation began at 9:00 a.m., and the competition continued until after 4:00 p.m. When it finally ended, I must admit that I had been so tense during the process that, several times after looking at the screen, my vision went black and I nearly fainted.

The panel of judges and consultants consisted of experts from various fields, and hearing that alone was enough to make anyone nervous. Today, I finally learned the results: in the intelligent video analysis competition, our company (ACTi Corporation) received a perfect score, while the other companies all lost points because of serious errors. The intelligent video analysis competition was only one part of the overall tender evaluation. Only those whose total scores across all competition items passed the threshold could qualify for the final tender competition. The results were as follows: Chunghwa Telecom scored 83, MiTAC scored 81, and Alcatel-Lucent (which adopted ACTi Corporation’s complete solution) scored 100. The other two companies lost their qualification. Although ACTi Corporation performed excellently and achieved a perfect score, the tender was not awarded solely on the basis of the highest score. In other words, technical excellence alone was not enough to secure the project, and the contract was ultimately awarded to another company. Our supervisors are now encouraging everyone and proposing solutions. I believe we will continue moving forward, because we are truly an outstanding team.

Figure 1. Our million-dollar surveillance vehicle. It is solar-chargeable and can independently supply power to four surveillance cameras. Its mast can be raised to as high as eight meters, giving it strong mobility. It is equipped with a trailer hitch so that it can be repositioned by towing. It is also protected against rain.

Figure 2. Our million-dollar solar-powered vehicle.

Figure 3. A photo of me with the solar-powered vehicle.

Figure 4. I was writing code by the roadside under the scorching sun.

Figure 5. These three sales engineers and application engineers were all outstanding teammates. They were not afraid of hard work, climbing up and down and enduring both wind and rain. Without them, I would not have been able to accomplish anything.

Figure 6. The site was already crowded several days before the competition.

Figure 7. This was another company’s lifting vehicle. It was quite fun, and it could even be driven around like a car.

Figure 8. Each company kept mounting its cameras higher and higher.

Figure 9. After the competition, the million-dollar solar-powered vehicle was towed away by a million-dollar BMW—a perfect match.

Author: Tai-Yu Lai, ChatGPT
Title: ACTi Receives a Perfect Score in the Taipei City Police Department Intelligent Surveillance System Tender Evaluation!